diff --git "a/examples/talk-llama/llama.cpp" "b/examples/talk-llama/llama.cpp" --- "a/examples/talk-llama/llama.cpp" +++ "b/examples/talk-llama/llama.cpp" @@ -1,48 +1,20 @@ #include "llama-impl.h" + +#include "llama-chat.h" +#include "llama-mmap.h" +#include "llama-context.h" #include "llama-vocab.h" #include "llama-sampling.h" - -#include "unicode.h" +#include "llama-kv-cache.h" +#include "llama-model-loader.h" +#include "llama-model.h" +#include "llama-quant.h" #include "ggml.h" #include "ggml-alloc.h" #include "ggml-backend.h" #include "ggml-cpp.h" -// TODO: replace with ggml API call -#define QK_K 256 - -#ifdef __has_include - #if __has_include() - #include - #if defined(_POSIX_MAPPED_FILES) - #include - #include - #endif - #if defined(_POSIX_MEMLOCK_RANGE) - #include - #endif - #endif -#endif - -#if defined(_WIN32) - #define WIN32_LEAN_AND_MEAN - #ifndef NOMINMAX - #define NOMINMAX - #endif - #include - #ifndef PATH_MAX - #define PATH_MAX MAX_PATH - #endif - #include -#endif - -#if __cplusplus >= 202000L - #define LU8(x) (const char*)(u8##x) -#else - #define LU8(x) u8##x -#endif - #include #include #include @@ -57,10865 +29,6338 @@ #include #include #include -#include #include -#include #include #include #include -#include -#include #include -#include -#include -#include #include -#include #if defined(_MSC_VER) #pragma warning(disable: 4244 4267) // possible loss of data #endif -// bump if necessary -#define LLAMA_MAX_LAYERS 512 -#define LLAMA_MAX_EXPERTS 160 // DeepSeekV2 - // -// helpers +// tensor loading (TODO: add llama_tesor_loader?) // -// trim whitespace from the beginning and end of a string -static std::string trim(const std::string & str) { - size_t start = 0; - size_t end = str.size(); - while (start < end && isspace(str[start])) { - start += 1; - } - while (end > start && isspace(str[end - 1])) { - end -= 1; - } - return str.substr(start, end - start); +static int llama_get_device_count(const llama_model & model) { + return (int) model.devices.size(); } -static bool is_float_close(float a, float b, float abs_tol) { - // Check for non-negative tolerance - if (abs_tol < 0.0) { - throw std::invalid_argument("Tolerance must be non-negative"); - } +// checks if the weight tensor can be used with the specified buffer type and device +static bool weight_buft_supported(const llama_hparams & hparams, ggml_tensor * w, ggml_op op, ggml_backend_buffer_type_t buft, ggml_backend_dev_t dev) { + GGML_ASSERT(w != nullptr); - // Exact equality check - if (a == b) { + if (op == GGML_OP_NONE) { return true; } - // Check for infinities - if (std::isinf(a) || std::isinf(b)) { - return false; + ggml_init_params params = { + /*.mem_size =*/ ggml_tensor_overhead()*8, + /*.mem_buffer =*/ NULL, + /*.no_alloc =*/ true, + }; + ggml_context_ptr ctx_ptr { ggml_init(params) }; + if (!ctx_ptr) { + throw std::runtime_error(format("failed to create ggml context")); } + ggml_context * ctx = ctx_ptr.get(); - // Regular comparison using the provided absolute tolerance - return std::fabs(b - a) <= abs_tol; -} + ggml_tensor * op_tensor = nullptr; + + switch (op) { + case GGML_OP_GET_ROWS: + { + ggml_tensor * b = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, 512); + op_tensor = ggml_get_rows(ctx, w, b); + } break; + case GGML_OP_MUL_MAT: + { + ggml_tensor * b = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, w->ne[0], 512, w->ne[2], w->ne[3]); + op_tensor = ggml_mul_mat(ctx, w, b); + } break; + case GGML_OP_MUL_MAT_ID: + { + int n_expert_used = hparams.n_expert_used; + ggml_tensor * b = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, w->ne[0], n_expert_used, 512); + ggml_tensor * ids = ggml_new_tensor_2d(ctx, GGML_TYPE_I32, n_expert_used, 512); + op_tensor = ggml_mul_mat_id(ctx, w, b, ids); + } break; + case GGML_OP_ADD: + { + ggml_tensor * a = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, w->ne[0], w->ne[1], w->ne[2], w->ne[3]); + op_tensor = ggml_add(ctx, a, w); + } break; + case GGML_OP_MUL: + { + ggml_tensor * a = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, w->ne[0], w->ne[1], w->ne[2], w->ne[3]); + op_tensor = ggml_mul(ctx, a, w); + } break; + case GGML_OP_DIV: + { + ggml_tensor * a = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, w->ne[0]); + op_tensor = ggml_div(ctx, a, w); + } break; + case GGML_OP_ROPE: + { + int n_embd_head = hparams.n_embd_head_v; + int n_head = hparams.n_head(); + ggml_tensor * a = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, n_embd_head, n_head, 512); + ggml_tensor * b = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, 512); + op_tensor = ggml_rope_ext( + ctx, a, b, w, + 0, 0, 0, 0, 0, + 0, 0, 0, 0 + ); -static void zeros(std::ofstream & file, size_t n) { - char zero = 0; - for (size_t i = 0; i < n; ++i) { - file.write(&zero, 1); + } break; + case GGML_OP_SSM_CONV: + { + // FIXME + ggml_tensor * conv_x = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, 12345, w->ne[1], 6789); + op_tensor = ggml_ssm_conv(ctx, conv_x, w); + } break; + case GGML_OP_SSM_SCAN: + { + // FIXME + const int64_t d_state = w->ne[0]; + const int64_t d_inner = w->ne[1]; + const int64_t n_seq_tokens = 512; + const int64_t n_seqs = 1; + ggml_tensor * s = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, d_state, d_inner, n_seqs); + ggml_tensor * x = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, d_inner, n_seq_tokens, n_seqs); + ggml_tensor * dt = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, d_inner, n_seq_tokens, n_seqs); + ggml_tensor * B = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, d_state, n_seq_tokens, n_seqs); + ggml_tensor * C = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, d_state, n_seq_tokens, n_seqs); + op_tensor = ggml_ssm_scan(ctx, s, x, dt, w, B, C); + } break; + case GGML_OP_RWKV_WKV6: + { + // FIXME + const int64_t S = 123; + const int64_t H = 123; + const int64_t n_tokens = 123; + const int64_t n_seqs = 123; + ggml_tensor * k = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, S, 1, H, n_tokens); + ggml_tensor * v = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, 1, S, H, n_tokens); + ggml_tensor * r = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, 1, S, H, n_tokens); + ggml_tensor * tf = w; + ggml_tensor * td = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, 1, S, H, n_tokens); + ggml_tensor * state = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, S, n_seqs, S, H); + op_tensor = ggml_rwkv_wkv6(ctx, k, v, r, tf, td, state); + } break; + case GGML_OP_IM2COL: + { + const int n_embd = hparams.n_embd; + ggml_tensor * b = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, n_embd, w->ne[1], 1, 1); + op_tensor = ggml_im2col(ctx, w, b, 1, 0, 0, 0, 1, 0, false, GGML_TYPE_F16); + } break; + default: + GGML_ABORT("%s: missing test for op %s for tensor %s", __func__, ggml_op_name(op), w->name); } + + // create a temporary dummy buffer for the weight so that supports_op can check the buffer type + GGML_ASSERT(w->buffer == nullptr); + w->buffer = ggml_backend_buft_alloc_buffer(buft, 0); + bool op_supported = ggml_backend_dev_supports_op(dev, op_tensor); + ggml_backend_buffer_free(w->buffer); + w->buffer = nullptr; + + return op_supported; } -LLAMA_ATTRIBUTE_FORMAT(1, 2) -static std::string format(const char * fmt, ...) { - va_list ap; - va_list ap2; - va_start(ap, fmt); - va_copy(ap2, ap); - int size = vsnprintf(NULL, 0, fmt, ap); - GGML_ASSERT(size >= 0 && size < INT_MAX); // NOLINT - std::vector buf(size + 1); - int size2 = vsnprintf(buf.data(), size + 1, fmt, ap2); - GGML_ASSERT(size2 == size); - va_end(ap2); - va_end(ap); - return std::string(buf.data(), size); +// find the first buffer type in the list that can use the tensor +static ggml_backend_buffer_type_t select_weight_buft(const llama_model & model, ggml_tensor * tensor, ggml_op op, const llama_model::buft_list_t & buft_list) { + GGML_ASSERT(!buft_list.empty()); + for (const auto & cur : buft_list) { + ggml_backend_dev_t cur_dev = cur.first; + ggml_backend_buffer_type_t cur_buft = cur.second; + if (weight_buft_supported(model.hparams, tensor, op, cur_buft, cur_dev)) { + return cur_buft; + } + } + return nullptr; } -// -// gguf constants (sync with gguf.py) -// +// CPU: ACCEL -> CPU extra -> GPU host -> CPU +static llama_model::buft_list_t make_cpu_buft_list(llama_model & model) { + llama_model::buft_list_t buft_list; -enum llm_arch { - LLM_ARCH_LLAMA, - LLM_ARCH_FALCON, - LLM_ARCH_BAICHUAN, - LLM_ARCH_GROK, - LLM_ARCH_GPT2, - LLM_ARCH_GPTJ, - LLM_ARCH_GPTNEOX, - LLM_ARCH_MPT, - LLM_ARCH_STARCODER, - LLM_ARCH_REFACT, - LLM_ARCH_BERT, - LLM_ARCH_NOMIC_BERT, - LLM_ARCH_JINA_BERT_V2, - LLM_ARCH_BLOOM, - LLM_ARCH_STABLELM, - LLM_ARCH_QWEN, - LLM_ARCH_QWEN2, - LLM_ARCH_QWEN2MOE, - LLM_ARCH_QWEN2VL, - LLM_ARCH_PHI2, - LLM_ARCH_PHI3, - LLM_ARCH_PLAMO, - LLM_ARCH_CODESHELL, - LLM_ARCH_ORION, - LLM_ARCH_INTERNLM2, - LLM_ARCH_MINICPM, - LLM_ARCH_MINICPM3, - LLM_ARCH_GEMMA, - LLM_ARCH_GEMMA2, - LLM_ARCH_STARCODER2, - LLM_ARCH_MAMBA, - LLM_ARCH_XVERSE, - LLM_ARCH_COMMAND_R, - LLM_ARCH_DBRX, - LLM_ARCH_OLMO, - LLM_ARCH_OLMO2, - LLM_ARCH_OLMOE, - LLM_ARCH_OPENELM, - LLM_ARCH_ARCTIC, - LLM_ARCH_DEEPSEEK, - LLM_ARCH_DEEPSEEK2, - LLM_ARCH_CHATGLM, - LLM_ARCH_BITNET, - LLM_ARCH_T5, - LLM_ARCH_T5ENCODER, - LLM_ARCH_JAIS, - LLM_ARCH_NEMOTRON, - LLM_ARCH_EXAONE, - LLM_ARCH_RWKV6, - LLM_ARCH_GRANITE, - LLM_ARCH_GRANITE_MOE, - LLM_ARCH_CHAMELEON, - LLM_ARCH_UNKNOWN, -}; + // add ACCEL buffer types + for (size_t i = 0; i < ggml_backend_dev_count(); ++i) { + ggml_backend_dev_t dev = ggml_backend_dev_get(i); + if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_ACCEL) { + auto * buft = ggml_backend_dev_buffer_type(dev); + // skip + if (buft != ggml_backend_cpu_buffer_type()) { + buft_list.emplace_back(dev, buft); + } + } + } -static const std::map LLM_ARCH_NAMES = { - { LLM_ARCH_LLAMA, "llama" }, - { LLM_ARCH_FALCON, "falcon" }, - { LLM_ARCH_GROK, "grok" }, - { LLM_ARCH_GPT2, "gpt2" }, - { LLM_ARCH_GPTJ, "gptj" }, - { LLM_ARCH_GPTNEOX, "gptneox" }, - { LLM_ARCH_MPT, "mpt" }, - { LLM_ARCH_BAICHUAN, "baichuan" }, - { LLM_ARCH_STARCODER, "starcoder" }, - { LLM_ARCH_REFACT, "refact" }, - { LLM_ARCH_BERT, "bert" }, - { LLM_ARCH_NOMIC_BERT, "nomic-bert" }, - { LLM_ARCH_JINA_BERT_V2, "jina-bert-v2" }, - { LLM_ARCH_BLOOM, "bloom" }, - { LLM_ARCH_STABLELM, "stablelm" }, - { LLM_ARCH_QWEN, "qwen" }, - { LLM_ARCH_QWEN2, "qwen2" }, - { LLM_ARCH_QWEN2MOE, "qwen2moe" }, - { LLM_ARCH_QWEN2VL, "qwen2vl" }, - { LLM_ARCH_PHI2, "phi2" }, - { LLM_ARCH_PHI3, "phi3" }, - { LLM_ARCH_PLAMO, "plamo" }, - { LLM_ARCH_CODESHELL, "codeshell" }, - { LLM_ARCH_ORION, "orion" }, - { LLM_ARCH_INTERNLM2, "internlm2" }, - { LLM_ARCH_MINICPM, "minicpm" }, - { LLM_ARCH_MINICPM3, "minicpm3" }, - { LLM_ARCH_GEMMA, "gemma" }, - { LLM_ARCH_GEMMA2, "gemma2" }, - { LLM_ARCH_STARCODER2, "starcoder2" }, - { LLM_ARCH_MAMBA, "mamba" }, - { LLM_ARCH_XVERSE, "xverse" }, - { LLM_ARCH_COMMAND_R, "command-r" }, - { LLM_ARCH_DBRX, "dbrx" }, - { LLM_ARCH_OLMO, "olmo" }, - { LLM_ARCH_OLMO2, "olmo2" }, - { LLM_ARCH_OLMOE, "olmoe" }, - { LLM_ARCH_OPENELM, "openelm" }, - { LLM_ARCH_ARCTIC, "arctic" }, - { LLM_ARCH_DEEPSEEK, "deepseek" }, - { LLM_ARCH_DEEPSEEK2, "deepseek2" }, - { LLM_ARCH_CHATGLM, "chatglm" }, - { LLM_ARCH_BITNET, "bitnet" }, - { LLM_ARCH_T5, "t5" }, - { LLM_ARCH_T5ENCODER, "t5encoder" }, - { LLM_ARCH_JAIS, "jais" }, - { LLM_ARCH_NEMOTRON, "nemotron" }, - { LLM_ARCH_EXAONE, "exaone" }, - { LLM_ARCH_RWKV6, "rwkv6" }, - { LLM_ARCH_GRANITE, "granite" }, - { LLM_ARCH_GRANITE_MOE, "granitemoe" }, - { LLM_ARCH_CHAMELEON, "chameleon" }, - { LLM_ARCH_UNKNOWN, "(unknown)" }, -}; + // add extra buffer types + auto * cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU); + auto * cpu_reg = ggml_backend_dev_backend_reg(cpu_dev); + auto ggml_backend_dev_get_extra_bufts_fn = (ggml_backend_dev_get_extra_bufts_t) + ggml_backend_reg_get_proc_address(cpu_reg, "ggml_backend_dev_get_extra_bufts"); + if (ggml_backend_dev_get_extra_bufts_fn) { + ggml_backend_buffer_type_t * extra_bufts = ggml_backend_dev_get_extra_bufts_fn(cpu_dev); + while (extra_bufts && *extra_bufts) { + buft_list.emplace_back(cpu_dev, *extra_bufts); + ++extra_bufts; + } + } -enum llm_kv { - LLM_KV_GENERAL_TYPE, - LLM_KV_GENERAL_ARCHITECTURE, - LLM_KV_GENERAL_QUANTIZATION_VERSION, - LLM_KV_GENERAL_ALIGNMENT, - LLM_KV_GENERAL_NAME, - LLM_KV_GENERAL_AUTHOR, - LLM_KV_GENERAL_VERSION, - LLM_KV_GENERAL_URL, - LLM_KV_GENERAL_DESCRIPTION, - LLM_KV_GENERAL_LICENSE, - LLM_KV_GENERAL_SOURCE_URL, - LLM_KV_GENERAL_SOURCE_HF_REPO, - - LLM_KV_VOCAB_SIZE, - LLM_KV_CONTEXT_LENGTH, - LLM_KV_EMBEDDING_LENGTH, - LLM_KV_BLOCK_COUNT, - LLM_KV_LEADING_DENSE_BLOCK_COUNT, - LLM_KV_FEED_FORWARD_LENGTH, - LLM_KV_EXPERT_FEED_FORWARD_LENGTH, - LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, - LLM_KV_USE_PARALLEL_RESIDUAL, - LLM_KV_TENSOR_DATA_LAYOUT, - LLM_KV_EXPERT_COUNT, - LLM_KV_EXPERT_USED_COUNT, - LLM_KV_EXPERT_SHARED_COUNT, - LLM_KV_EXPERT_WEIGHTS_SCALE, - LLM_KV_POOLING_TYPE, - LLM_KV_LOGIT_SCALE, - LLM_KV_DECODER_START_TOKEN_ID, - LLM_KV_ATTN_LOGIT_SOFTCAPPING, - LLM_KV_FINAL_LOGIT_SOFTCAPPING, - LLM_KV_SWIN_NORM, - LLM_KV_RESCALE_EVERY_N_LAYERS, - LLM_KV_TIME_MIX_EXTRA_DIM, - LLM_KV_TIME_DECAY_EXTRA_DIM, - LLM_KV_RESIDUAL_SCALE, - LLM_KV_EMBEDDING_SCALE, - - LLM_KV_ATTENTION_HEAD_COUNT, - LLM_KV_ATTENTION_HEAD_COUNT_KV, - LLM_KV_ATTENTION_MAX_ALIBI_BIAS, - LLM_KV_ATTENTION_CLAMP_KQV, - LLM_KV_ATTENTION_KEY_LENGTH, - LLM_KV_ATTENTION_VALUE_LENGTH, - LLM_KV_ATTENTION_LAYERNORM_EPS, - LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, - LLM_KV_ATTENTION_CAUSAL, - LLM_KV_ATTENTION_Q_LORA_RANK, - LLM_KV_ATTENTION_KV_LORA_RANK, - LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT, - LLM_KV_ATTENTION_SLIDING_WINDOW, - LLM_KV_ATTENTION_SCALE, - - LLM_KV_ROPE_DIMENSION_COUNT, - LLM_KV_ROPE_DIMENSION_SECTIONS, - LLM_KV_ROPE_FREQ_BASE, - LLM_KV_ROPE_SCALE_LINEAR, - LLM_KV_ROPE_SCALING_TYPE, - LLM_KV_ROPE_SCALING_FACTOR, - LLM_KV_ROPE_SCALING_ATTN_FACTOR, - LLM_KV_ROPE_SCALING_ORIG_CTX_LEN, - LLM_KV_ROPE_SCALING_FINETUNED, - LLM_KV_ROPE_SCALING_YARN_LOG_MUL, - - LLM_KV_SPLIT_NO, - LLM_KV_SPLIT_COUNT, - LLM_KV_SPLIT_TENSORS_COUNT, - - LLM_KV_SSM_INNER_SIZE, - LLM_KV_SSM_CONV_KERNEL, - LLM_KV_SSM_STATE_SIZE, - LLM_KV_SSM_TIME_STEP_RANK, - LLM_KV_SSM_DT_B_C_RMS, - - LLM_KV_WKV_HEAD_SIZE, - - LLM_KV_TOKENIZER_MODEL, - LLM_KV_TOKENIZER_PRE, - LLM_KV_TOKENIZER_LIST, - LLM_KV_TOKENIZER_TOKEN_TYPE, - LLM_KV_TOKENIZER_TOKEN_TYPE_COUNT, - LLM_KV_TOKENIZER_SCORES, - LLM_KV_TOKENIZER_MERGES, - LLM_KV_TOKENIZER_BOS_ID, - LLM_KV_TOKENIZER_EOS_ID, - LLM_KV_TOKENIZER_EOT_ID, - LLM_KV_TOKENIZER_EOM_ID, - LLM_KV_TOKENIZER_UNK_ID, - LLM_KV_TOKENIZER_SEP_ID, - LLM_KV_TOKENIZER_PAD_ID, - LLM_KV_TOKENIZER_CLS_ID, - LLM_KV_TOKENIZER_MASK_ID, - LLM_KV_TOKENIZER_ADD_BOS, - LLM_KV_TOKENIZER_ADD_EOS, - LLM_KV_TOKENIZER_ADD_PREFIX, - LLM_KV_TOKENIZER_REMOVE_EXTRA_WS, - LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP, - LLM_KV_TOKENIZER_HF_JSON, - LLM_KV_TOKENIZER_RWKV, - LLM_KV_TOKENIZER_FIM_PRE_ID, - LLM_KV_TOKENIZER_FIM_SUF_ID, - LLM_KV_TOKENIZER_FIM_MID_ID, - LLM_KV_TOKENIZER_FIM_PAD_ID, - LLM_KV_TOKENIZER_FIM_REP_ID, - LLM_KV_TOKENIZER_FIM_SEP_ID, - - LLM_KV_ADAPTER_TYPE, - LLM_KV_ADAPTER_LORA_ALPHA, - - // deprecated: - LLM_KV_TOKENIZER_PREFIX_ID, - LLM_KV_TOKENIZER_SUFFIX_ID, - LLM_KV_TOKENIZER_MIDDLE_ID, -}; + // add a host buffer type + // storing the tensors in a host buffer is useful when the processing of large batches + // is offloaded to a GPU device, since it reduces the time spent on data transfers + // generally, this will be done using the first device in the list + // a better approach would be to handle this on a weight-by-weight basis using the offload_op + // function of the device to determine if it would benefit from being stored in a host buffer + for (auto * dev : model.devices) { + ggml_backend_buffer_type_t buft = ggml_backend_dev_host_buffer_type(dev); + if (buft) { + buft_list.emplace_back(dev, buft); + break; + } + } -static const std::map LLM_KV_NAMES = { - { LLM_KV_GENERAL_TYPE, "general.type" }, - { LLM_KV_GENERAL_ARCHITECTURE, "general.architecture" }, - { LLM_KV_GENERAL_QUANTIZATION_VERSION, "general.quantization_version" }, - { LLM_KV_GENERAL_ALIGNMENT, "general.alignment" }, - { LLM_KV_GENERAL_NAME, "general.name" }, - { LLM_KV_GENERAL_AUTHOR, "general.author" }, - { LLM_KV_GENERAL_VERSION, "general.version" }, - { LLM_KV_GENERAL_URL, "general.url" }, - { LLM_KV_GENERAL_DESCRIPTION, "general.description" }, - { LLM_KV_GENERAL_LICENSE, "general.license" }, - { LLM_KV_GENERAL_SOURCE_URL, "general.source.url" }, - { LLM_KV_GENERAL_SOURCE_HF_REPO, "general.source.huggingface.repository" }, - - { LLM_KV_VOCAB_SIZE, "%s.vocab_size" }, - { LLM_KV_CONTEXT_LENGTH, "%s.context_length" }, - { LLM_KV_EMBEDDING_LENGTH, "%s.embedding_length" }, - { LLM_KV_BLOCK_COUNT, "%s.block_count" }, - { LLM_KV_LEADING_DENSE_BLOCK_COUNT, "%s.leading_dense_block_count" }, - { LLM_KV_FEED_FORWARD_LENGTH, "%s.feed_forward_length" }, - { LLM_KV_EXPERT_FEED_FORWARD_LENGTH, "%s.expert_feed_forward_length" }, - { LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, "%s.expert_shared_feed_forward_length" }, - { LLM_KV_USE_PARALLEL_RESIDUAL, "%s.use_parallel_residual" }, - { LLM_KV_TENSOR_DATA_LAYOUT, "%s.tensor_data_layout" }, - { LLM_KV_EXPERT_COUNT, "%s.expert_count" }, - { LLM_KV_EXPERT_USED_COUNT, "%s.expert_used_count" }, - { LLM_KV_EXPERT_SHARED_COUNT, "%s.expert_shared_count" }, - { LLM_KV_EXPERT_WEIGHTS_SCALE, "%s.expert_weights_scale" }, - { LLM_KV_POOLING_TYPE, "%s.pooling_type" }, - { LLM_KV_LOGIT_SCALE, "%s.logit_scale" }, - { LLM_KV_DECODER_START_TOKEN_ID, "%s.decoder_start_token_id" }, - { LLM_KV_ATTN_LOGIT_SOFTCAPPING, "%s.attn_logit_softcapping" }, - { LLM_KV_FINAL_LOGIT_SOFTCAPPING, "%s.final_logit_softcapping" }, - { LLM_KV_SWIN_NORM, "%s.swin_norm" }, - { LLM_KV_RESCALE_EVERY_N_LAYERS, "%s.rescale_every_n_layers" }, - { LLM_KV_TIME_MIX_EXTRA_DIM, "%s.time_mix_extra_dim" }, - { LLM_KV_TIME_DECAY_EXTRA_DIM, "%s.time_decay_extra_dim" }, - { LLM_KV_RESIDUAL_SCALE, "%s.residual_scale" }, - { LLM_KV_EMBEDDING_SCALE, "%s.embedding_scale" }, - - { LLM_KV_ATTENTION_HEAD_COUNT, "%s.attention.head_count" }, - { LLM_KV_ATTENTION_HEAD_COUNT_KV, "%s.attention.head_count_kv" }, - { LLM_KV_ATTENTION_MAX_ALIBI_BIAS, "%s.attention.max_alibi_bias" }, - { LLM_KV_ATTENTION_CLAMP_KQV, "%s.attention.clamp_kqv" }, - { LLM_KV_ATTENTION_KEY_LENGTH, "%s.attention.key_length" }, - { LLM_KV_ATTENTION_VALUE_LENGTH, "%s.attention.value_length" }, - { LLM_KV_ATTENTION_LAYERNORM_EPS, "%s.attention.layer_norm_epsilon" }, - { LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, "%s.attention.layer_norm_rms_epsilon" }, - { LLM_KV_ATTENTION_CAUSAL, "%s.attention.causal" }, - { LLM_KV_ATTENTION_Q_LORA_RANK, "%s.attention.q_lora_rank" }, - { LLM_KV_ATTENTION_KV_LORA_RANK, "%s.attention.kv_lora_rank" }, - { LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT, "%s.attention.relative_buckets_count" }, - { LLM_KV_ATTENTION_SLIDING_WINDOW, "%s.attention.sliding_window" }, - { LLM_KV_ATTENTION_SCALE, "%s.attention.scale" }, - - { LLM_KV_ROPE_DIMENSION_COUNT, "%s.rope.dimension_count" }, - { LLM_KV_ROPE_DIMENSION_SECTIONS, "%s.rope.dimension_sections" }, - { LLM_KV_ROPE_FREQ_BASE, "%s.rope.freq_base" }, - { LLM_KV_ROPE_SCALE_LINEAR, "%s.rope.scale_linear" }, - { LLM_KV_ROPE_SCALING_TYPE, "%s.rope.scaling.type" }, - { LLM_KV_ROPE_SCALING_FACTOR, "%s.rope.scaling.factor" }, - { LLM_KV_ROPE_SCALING_ATTN_FACTOR, "%s.rope.scaling.attn_factor" }, - { LLM_KV_ROPE_SCALING_ORIG_CTX_LEN, "%s.rope.scaling.original_context_length" }, - { LLM_KV_ROPE_SCALING_FINETUNED, "%s.rope.scaling.finetuned" }, - { LLM_KV_ROPE_SCALING_YARN_LOG_MUL, "%s.rope.scaling.yarn_log_multiplier" }, - - { LLM_KV_SPLIT_NO, "split.no" }, - { LLM_KV_SPLIT_COUNT, "split.count" }, - { LLM_KV_SPLIT_TENSORS_COUNT, "split.tensors.count" }, - - { LLM_KV_SSM_CONV_KERNEL, "%s.ssm.conv_kernel" }, - { LLM_KV_SSM_INNER_SIZE, "%s.ssm.inner_size" }, - { LLM_KV_SSM_STATE_SIZE, "%s.ssm.state_size" }, - { LLM_KV_SSM_TIME_STEP_RANK, "%s.ssm.time_step_rank" }, - { LLM_KV_SSM_DT_B_C_RMS, "%s.ssm.dt_b_c_rms" }, - - { LLM_KV_WKV_HEAD_SIZE, "%s.wkv.head_size" }, - - { LLM_KV_TOKENIZER_MODEL, "tokenizer.ggml.model" }, - { LLM_KV_TOKENIZER_PRE, "tokenizer.ggml.pre" }, - { LLM_KV_TOKENIZER_LIST, "tokenizer.ggml.tokens" }, - { LLM_KV_TOKENIZER_TOKEN_TYPE, "tokenizer.ggml.token_type" }, - { LLM_KV_TOKENIZER_TOKEN_TYPE_COUNT, "tokenizer.ggml.token_type_count" }, - { LLM_KV_TOKENIZER_SCORES, "tokenizer.ggml.scores" }, - { LLM_KV_TOKENIZER_MERGES, "tokenizer.ggml.merges" }, - { LLM_KV_TOKENIZER_BOS_ID, "tokenizer.ggml.bos_token_id" }, - { LLM_KV_TOKENIZER_EOS_ID, "tokenizer.ggml.eos_token_id" }, - { LLM_KV_TOKENIZER_EOT_ID, "tokenizer.ggml.eot_token_id" }, - { LLM_KV_TOKENIZER_EOM_ID, "tokenizer.ggml.eom_token_id" }, - { LLM_KV_TOKENIZER_UNK_ID, "tokenizer.ggml.unknown_token_id" }, - { LLM_KV_TOKENIZER_SEP_ID, "tokenizer.ggml.seperator_token_id" }, - { LLM_KV_TOKENIZER_PAD_ID, "tokenizer.ggml.padding_token_id" }, - { LLM_KV_TOKENIZER_CLS_ID, "tokenizer.ggml.cls_token_id" }, - { LLM_KV_TOKENIZER_MASK_ID, "tokenizer.ggml.mask_token_id" }, - { LLM_KV_TOKENIZER_ADD_BOS, "tokenizer.ggml.add_bos_token" }, - { LLM_KV_TOKENIZER_ADD_EOS, "tokenizer.ggml.add_eos_token" }, - { LLM_KV_TOKENIZER_ADD_PREFIX, "tokenizer.ggml.add_space_prefix" }, - { LLM_KV_TOKENIZER_REMOVE_EXTRA_WS, "tokenizer.ggml.remove_extra_whitespaces" }, - { LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP, "tokenizer.ggml.precompiled_charsmap" }, - { LLM_KV_TOKENIZER_HF_JSON, "tokenizer.huggingface.json" }, - { LLM_KV_TOKENIZER_RWKV, "tokenizer.rwkv.world" }, - { LLM_KV_TOKENIZER_FIM_PRE_ID, "tokenizer.ggml.fim_pre_token_id" }, - { LLM_KV_TOKENIZER_FIM_SUF_ID, "tokenizer.ggml.fim_suf_token_id" }, - { LLM_KV_TOKENIZER_FIM_MID_ID, "tokenizer.ggml.fim_mid_token_id" }, - { LLM_KV_TOKENIZER_FIM_PAD_ID, "tokenizer.ggml.fim_pad_token_id" }, - { LLM_KV_TOKENIZER_FIM_REP_ID, "tokenizer.ggml.fim_rep_token_id" }, - { LLM_KV_TOKENIZER_FIM_SEP_ID, "tokenizer.ggml.fim_sep_token_id" }, - - { LLM_KV_ADAPTER_TYPE, "adapter.type" }, - { LLM_KV_ADAPTER_LORA_ALPHA, "adapter.lora.alpha" }, - - // deprecated - { LLM_KV_TOKENIZER_PREFIX_ID, "tokenizer.ggml.prefix_token_id" }, - { LLM_KV_TOKENIZER_SUFFIX_ID, "tokenizer.ggml.suffix_token_id" }, - { LLM_KV_TOKENIZER_MIDDLE_ID, "tokenizer.ggml.middle_token_id" }, -}; + // add the CPU buffer type + for (size_t i = 0; i < ggml_backend_dev_count(); ++i) { + ggml_backend_dev_t dev = ggml_backend_dev_get(i); + if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_CPU) { + buft_list.emplace_back(dev, ggml_backend_dev_buffer_type(dev)); + } + } -struct LLM_KV { - LLM_KV(llm_arch arch) : arch(arch) {} + return buft_list; +} - llm_arch arch; +// GPU: split if LLAMA_SPLIT_MODE_ROW -> GPU +static llama_model::buft_list_t make_gpu_buft_list(ggml_backend_dev_t dev, enum llama_split_mode split_mode, const float * tensor_split) { + llama_model::buft_list_t buft_list; - std::string operator()(llm_kv kv) const { - return ::format(LLM_KV_NAMES.at(kv), LLM_ARCH_NAMES.at(arch)); + // add the device split buffer type if requested and available + if (split_mode == LLAMA_SPLIT_MODE_ROW) { + ggml_backend_reg_t reg = ggml_backend_dev_backend_reg(dev); + auto ggml_backend_split_buffer_type_fn = (ggml_backend_split_buffer_type_t) + ggml_backend_reg_get_proc_address(reg, "ggml_backend_split_buffer_type"); + if (ggml_backend_split_buffer_type_fn) { + size_t dev_index = [&]() { + auto * reg = ggml_backend_dev_backend_reg(dev); + for (size_t i = 0; i < ggml_backend_reg_dev_count(reg); ++i) { + if (ggml_backend_reg_dev_get(reg, i) == dev) { + return i; + } + } + throw std::runtime_error(format("device %s not found in its backend reg", ggml_backend_dev_name(dev))); + }(); + auto * buft = ggml_backend_split_buffer_type_fn(dev_index, tensor_split); + if (buft != nullptr) { + buft_list.emplace_back(dev, buft); + } + } } -}; -enum llm_tensor { - LLM_TENSOR_TOKEN_EMBD, - LLM_TENSOR_TOKEN_EMBD_NORM, - LLM_TENSOR_TOKEN_TYPES, - LLM_TENSOR_POS_EMBD, - LLM_TENSOR_OUTPUT, - LLM_TENSOR_OUTPUT_NORM, - LLM_TENSOR_ROPE_FREQS, - LLM_TENSOR_ROPE_FACTORS_LONG, - LLM_TENSOR_ROPE_FACTORS_SHORT, - LLM_TENSOR_ATTN_Q, - LLM_TENSOR_ATTN_K, - LLM_TENSOR_ATTN_V, - LLM_TENSOR_ATTN_QKV, - LLM_TENSOR_ATTN_OUT, - LLM_TENSOR_ATTN_NORM, - LLM_TENSOR_ATTN_NORM_2, - LLM_TENSOR_ATTN_OUT_NORM, - LLM_TENSOR_ATTN_POST_NORM, - LLM_TENSOR_ATTN_ROT_EMBD, - LLM_TENSOR_FFN_GATE_INP, - LLM_TENSOR_FFN_GATE_INP_SHEXP, - LLM_TENSOR_FFN_NORM, - LLM_TENSOR_FFN_POST_NORM, - LLM_TENSOR_FFN_GATE, - LLM_TENSOR_FFN_DOWN, - LLM_TENSOR_FFN_UP, - LLM_TENSOR_FFN_ACT, - LLM_TENSOR_FFN_DOWN_EXP, // split experts for backward compatibility - LLM_TENSOR_FFN_GATE_EXP, - LLM_TENSOR_FFN_UP_EXP, - LLM_TENSOR_FFN_NORM_EXPS, - LLM_TENSOR_FFN_DOWN_EXPS, // merged experts - LLM_TENSOR_FFN_GATE_EXPS, - LLM_TENSOR_FFN_UP_EXPS, - LLM_TENSOR_FFN_DOWN_SHEXP, - LLM_TENSOR_FFN_GATE_SHEXP, - LLM_TENSOR_FFN_UP_SHEXP, - LLM_TENSOR_ATTN_Q_NORM, - LLM_TENSOR_ATTN_K_NORM, - LLM_TENSOR_LAYER_OUT_NORM, - LLM_TENSOR_SSM_IN, - LLM_TENSOR_SSM_CONV1D, - LLM_TENSOR_SSM_X, - LLM_TENSOR_SSM_DT, - LLM_TENSOR_SSM_A, - LLM_TENSOR_SSM_D, - LLM_TENSOR_SSM_OUT, - LLM_TENSOR_TIME_MIX_W1, - LLM_TENSOR_TIME_MIX_W2, - LLM_TENSOR_TIME_MIX_LERP_X, - LLM_TENSOR_TIME_MIX_LERP_W, - LLM_TENSOR_TIME_MIX_LERP_K, - LLM_TENSOR_TIME_MIX_LERP_V, - LLM_TENSOR_TIME_MIX_LERP_R, - LLM_TENSOR_TIME_MIX_LERP_G, - LLM_TENSOR_TIME_MIX_FIRST, - LLM_TENSOR_TIME_MIX_DECAY, - LLM_TENSOR_TIME_MIX_DECAY_W1, - LLM_TENSOR_TIME_MIX_DECAY_W2, - LLM_TENSOR_TIME_MIX_KEY, - LLM_TENSOR_TIME_MIX_VALUE, - LLM_TENSOR_TIME_MIX_RECEPTANCE, - LLM_TENSOR_TIME_MIX_GATE, - LLM_TENSOR_TIME_MIX_LN, - LLM_TENSOR_TIME_MIX_OUTPUT, - LLM_TENSOR_CHANNEL_MIX_LERP_K, - LLM_TENSOR_CHANNEL_MIX_LERP_R, - LLM_TENSOR_CHANNEL_MIX_KEY, - LLM_TENSOR_CHANNEL_MIX_RECEPTANCE, - LLM_TENSOR_CHANNEL_MIX_VALUE, - LLM_TENSOR_ATTN_Q_A, - LLM_TENSOR_ATTN_Q_B, - LLM_TENSOR_ATTN_KV_A_MQA, - LLM_TENSOR_ATTN_KV_B, - LLM_TENSOR_ATTN_Q_A_NORM, - LLM_TENSOR_ATTN_KV_A_NORM, - LLM_TENSOR_ATTN_SUB_NORM, - LLM_TENSOR_FFN_SUB_NORM, - LLM_TENSOR_DEC_ATTN_NORM, - LLM_TENSOR_DEC_ATTN_Q, - LLM_TENSOR_DEC_ATTN_K, - LLM_TENSOR_DEC_ATTN_V, - LLM_TENSOR_DEC_ATTN_OUT, - LLM_TENSOR_DEC_ATTN_REL_B, - LLM_TENSOR_DEC_CROSS_ATTN_NORM, - LLM_TENSOR_DEC_CROSS_ATTN_Q, - LLM_TENSOR_DEC_CROSS_ATTN_K, - LLM_TENSOR_DEC_CROSS_ATTN_V, - LLM_TENSOR_DEC_CROSS_ATTN_OUT, - LLM_TENSOR_DEC_CROSS_ATTN_REL_B, - LLM_TENSOR_DEC_FFN_NORM, - LLM_TENSOR_DEC_FFN_GATE, - LLM_TENSOR_DEC_FFN_DOWN, - LLM_TENSOR_DEC_FFN_UP, - LLM_TENSOR_DEC_OUTPUT_NORM, - LLM_TENSOR_ENC_ATTN_NORM, - LLM_TENSOR_ENC_ATTN_Q, - LLM_TENSOR_ENC_ATTN_K, - LLM_TENSOR_ENC_ATTN_V, - LLM_TENSOR_ENC_ATTN_OUT, - LLM_TENSOR_ENC_ATTN_REL_B, - LLM_TENSOR_ENC_FFN_NORM, - LLM_TENSOR_ENC_FFN_GATE, - LLM_TENSOR_ENC_FFN_DOWN, - LLM_TENSOR_ENC_FFN_UP, - LLM_TENSOR_ENC_OUTPUT_NORM, - LLM_TENSOR_CLS, - LLM_TENSOR_CLS_OUT, -}; + // add the device default buffer type + buft_list.emplace_back(dev, ggml_backend_dev_buffer_type(dev)); -static const std::map> LLM_TENSOR_NAMES = { - { - LLM_ARCH_LLAMA, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" }, - { LLM_TENSOR_FFN_GATE_INP, "blk.%d.ffn_gate_inp" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_GATE_EXP, "blk.%d.ffn_gate.%d" }, - { LLM_TENSOR_FFN_DOWN_EXP, "blk.%d.ffn_down.%d" }, - { LLM_TENSOR_FFN_UP_EXP, "blk.%d.ffn_up.%d" }, - { LLM_TENSOR_FFN_GATE_EXPS, "blk.%d.ffn_gate_exps" }, - { LLM_TENSOR_FFN_DOWN_EXPS, "blk.%d.ffn_down_exps" }, - { LLM_TENSOR_FFN_UP_EXPS, "blk.%d.ffn_up_exps" }, - }, - }, - { - LLM_ARCH_BAICHUAN, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_FALCON, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_NORM_2, "blk.%d.attn_norm_2" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_GROK, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" }, - { LLM_TENSOR_FFN_GATE_INP, "blk.%d.ffn_gate_inp" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE_EXP, "blk.%d.ffn_gate.%d" }, - { LLM_TENSOR_FFN_DOWN_EXP, "blk.%d.ffn_down.%d" }, - { LLM_TENSOR_FFN_UP_EXP, "blk.%d.ffn_up.%d" }, - { LLM_TENSOR_FFN_GATE_EXPS, "blk.%d.ffn_gate_exps" }, - { LLM_TENSOR_FFN_DOWN_EXPS, "blk.%d.ffn_down_exps" }, - { LLM_TENSOR_FFN_UP_EXPS, "blk.%d.ffn_up_exps" }, - { LLM_TENSOR_LAYER_OUT_NORM, "blk.%d.layer_output_norm" }, - { LLM_TENSOR_ATTN_OUT_NORM, "blk.%d.attn_output_norm" }, - }, - }, - { - LLM_ARCH_GPT2, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_POS_EMBD, "position_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - }, - }, - { - LLM_ARCH_GPTJ, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - }, - }, - { - LLM_ARCH_GPTNEOX, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_MPT, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output"}, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_ACT, "blk.%d.ffn.act" }, - { LLM_TENSOR_POS_EMBD, "position_embd" }, - { LLM_TENSOR_ATTN_Q_NORM, "blk.%d.attn_q_norm"}, - { LLM_TENSOR_ATTN_K_NORM, "blk.%d.attn_k_norm"}, - }, - }, - { - LLM_ARCH_STARCODER, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_POS_EMBD, "position_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - }, - }, - { - LLM_ARCH_REFACT, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_BERT, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" }, - { LLM_TENSOR_TOKEN_TYPES, "token_types" }, - { LLM_TENSOR_POS_EMBD, "position_embd" }, - { LLM_TENSOR_ATTN_OUT_NORM, "blk.%d.attn_output_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_LAYER_OUT_NORM, "blk.%d.layer_output_norm" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_CLS, "cls" }, - { LLM_TENSOR_CLS_OUT, "cls.output" }, - }, - }, - { - LLM_ARCH_NOMIC_BERT, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" }, - { LLM_TENSOR_TOKEN_TYPES, "token_types" }, - { LLM_TENSOR_ATTN_OUT_NORM, "blk.%d.attn_output_norm" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_LAYER_OUT_NORM, "blk.%d.layer_output_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_JINA_BERT_V2, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" }, - { LLM_TENSOR_TOKEN_TYPES, "token_types" }, - { LLM_TENSOR_ATTN_NORM_2, "blk.%d.attn_norm_2" }, - { LLM_TENSOR_ATTN_OUT_NORM, "blk.%d.attn_output_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_Q_NORM, "blk.%d.attn_q_norm" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_K_NORM, "blk.%d.attn_k_norm" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_LAYER_OUT_NORM, "blk.%d.layer_output_norm" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_CLS, "cls" }, - }, - }, - { - LLM_ARCH_BLOOM, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - }, - }, - { - LLM_ARCH_STABLELM, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_ATTN_Q_NORM, "blk.%d.attn_q_norm" }, - { LLM_TENSOR_ATTN_K_NORM, "blk.%d.attn_k_norm" }, - }, - }, - { - LLM_ARCH_QWEN, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_QWEN2, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_QWEN2VL, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_QWEN2MOE, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE_INP, "blk.%d.ffn_gate_inp" }, - { LLM_TENSOR_FFN_GATE_EXPS, "blk.%d.ffn_gate_exps" }, - { LLM_TENSOR_FFN_DOWN_EXPS, "blk.%d.ffn_down_exps" }, - { LLM_TENSOR_FFN_UP_EXPS, "blk.%d.ffn_up_exps" }, - { LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" }, - { LLM_TENSOR_FFN_GATE_SHEXP, "blk.%d.ffn_gate_shexp" }, - { LLM_TENSOR_FFN_DOWN_SHEXP, "blk.%d.ffn_down_shexp" }, - { LLM_TENSOR_FFN_UP_SHEXP, "blk.%d.ffn_up_shexp" }, - }, - }, - { - LLM_ARCH_PHI2, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_PHI3, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FACTORS_LONG, "rope_factors_long" }, - { LLM_TENSOR_ROPE_FACTORS_SHORT, "rope_factors_short" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_PLAMO, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_CODESHELL, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_ORION, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_INTERNLM2, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_MINICPM, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ROPE_FACTORS_LONG, "rope_factors_long" }, - { LLM_TENSOR_ROPE_FACTORS_SHORT, "rope_factors_short" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" }, - { LLM_TENSOR_FFN_GATE_INP, "blk.%d.ffn_gate_inp" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_GATE_EXP, "blk.%d.ffn_gate.%d" }, - { LLM_TENSOR_FFN_DOWN_EXP, "blk.%d.ffn_down.%d" }, - { LLM_TENSOR_FFN_UP_EXP, "blk.%d.ffn_up.%d" }, - }, - }, - { - LLM_ARCH_MINICPM3, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FACTORS_LONG, "rope_factors_long" }, - { LLM_TENSOR_ROPE_FACTORS_SHORT, "rope_factors_short" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q_A_NORM, "blk.%d.attn_q_a_norm" }, - { LLM_TENSOR_ATTN_KV_A_NORM, "blk.%d.attn_kv_a_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_Q_A, "blk.%d.attn_q_a" }, - { LLM_TENSOR_ATTN_Q_B, "blk.%d.attn_q_b" }, - { LLM_TENSOR_ATTN_KV_A_MQA, "blk.%d.attn_kv_a_mqa" }, - { LLM_TENSOR_ATTN_KV_B, "blk.%d.attn_kv_b" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - }, - }, - { - LLM_ARCH_GEMMA, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_GEMMA2, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_POST_NORM, "blk.%d.post_attention_norm" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_POST_NORM, "blk.%d.post_ffw_norm" }, - }, - }, - { - LLM_ARCH_STARCODER2, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_MAMBA, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_SSM_IN, "blk.%d.ssm_in" }, - { LLM_TENSOR_SSM_CONV1D, "blk.%d.ssm_conv1d" }, - { LLM_TENSOR_SSM_X, "blk.%d.ssm_x" }, - { LLM_TENSOR_SSM_DT, "blk.%d.ssm_dt" }, - { LLM_TENSOR_SSM_A, "blk.%d.ssm_a" }, - { LLM_TENSOR_SSM_D, "blk.%d.ssm_d" }, - { LLM_TENSOR_SSM_OUT, "blk.%d.ssm_out" }, - }, - }, - { - LLM_ARCH_XVERSE, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_COMMAND_R, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_ATTN_Q_NORM, "blk.%d.attn_q_norm" }, - { LLM_TENSOR_ATTN_K_NORM, "blk.%d.attn_k_norm" }, - }, - }, - { - LLM_ARCH_DBRX, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_OUT_NORM, "blk.%d.attn_output_norm" }, - { LLM_TENSOR_FFN_GATE_INP, "blk.%d.ffn_gate_inp" }, - { LLM_TENSOR_FFN_GATE_EXPS, "blk.%d.ffn_gate_exps" }, - { LLM_TENSOR_FFN_DOWN_EXPS, "blk.%d.ffn_down_exps" }, - { LLM_TENSOR_FFN_UP_EXPS, "blk.%d.ffn_up_exps" }, - }, - }, - { - LLM_ARCH_OLMO, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_OLMO2, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_POST_NORM, "blk.%d.post_attention_norm" }, - { LLM_TENSOR_ATTN_Q_NORM, "blk.%d.attn_q_norm" }, - { LLM_TENSOR_ATTN_K_NORM, "blk.%d.attn_k_norm" }, - { LLM_TENSOR_FFN_POST_NORM, "blk.%d.post_ffw_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_OLMOE, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_Q_NORM, "blk.%d.attn_q_norm" }, - { LLM_TENSOR_ATTN_K_NORM, "blk.%d.attn_k_norm" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE_INP, "blk.%d.ffn_gate_inp" }, - { LLM_TENSOR_FFN_GATE_EXPS, "blk.%d.ffn_gate_exps" }, - { LLM_TENSOR_FFN_DOWN_EXPS, "blk.%d.ffn_down_exps" }, - { LLM_TENSOR_FFN_UP_EXPS, "blk.%d.ffn_up_exps" }, - }, - }, - { - LLM_ARCH_OPENELM, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_Q_NORM, "blk.%d.attn_q_norm" }, - { LLM_TENSOR_ATTN_K_NORM, "blk.%d.attn_k_norm" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_ARCTIC, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_GATE_INP, "blk.%d.ffn_gate_inp" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_NORM_EXPS, "blk.%d.ffn_norm_exps" }, - { LLM_TENSOR_FFN_GATE_EXPS, "blk.%d.ffn_gate_exps" }, - { LLM_TENSOR_FFN_DOWN_EXPS, "blk.%d.ffn_down_exps" }, - { LLM_TENSOR_FFN_UP_EXPS, "blk.%d.ffn_up_exps" }, - }, - }, - { - LLM_ARCH_DEEPSEEK, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" }, - { LLM_TENSOR_FFN_GATE_INP, "blk.%d.ffn_gate_inp" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_GATE_EXPS, "blk.%d.ffn_gate_exps" }, - { LLM_TENSOR_FFN_DOWN_EXPS, "blk.%d.ffn_down_exps" }, - { LLM_TENSOR_FFN_UP_EXPS, "blk.%d.ffn_up_exps" }, - { LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" }, - { LLM_TENSOR_FFN_GATE_SHEXP, "blk.%d.ffn_gate_shexp" }, - { LLM_TENSOR_FFN_DOWN_SHEXP, "blk.%d.ffn_down_shexp" }, - { LLM_TENSOR_FFN_UP_SHEXP, "blk.%d.ffn_up_shexp" }, - }, - }, - { - LLM_ARCH_DEEPSEEK2, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q_A_NORM, "blk.%d.attn_q_a_norm" }, - { LLM_TENSOR_ATTN_KV_A_NORM, "blk.%d.attn_kv_a_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_Q_A, "blk.%d.attn_q_a" }, - { LLM_TENSOR_ATTN_Q_B, "blk.%d.attn_q_b" }, - { LLM_TENSOR_ATTN_KV_A_MQA, "blk.%d.attn_kv_a_mqa" }, - { LLM_TENSOR_ATTN_KV_B, "blk.%d.attn_kv_b" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_GATE_INP, "blk.%d.ffn_gate_inp" }, - { LLM_TENSOR_FFN_GATE_EXPS, "blk.%d.ffn_gate_exps" }, - { LLM_TENSOR_FFN_DOWN_EXPS, "blk.%d.ffn_down_exps" }, - { LLM_TENSOR_FFN_UP_EXPS, "blk.%d.ffn_up_exps" }, - { LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" }, - { LLM_TENSOR_FFN_GATE_SHEXP, "blk.%d.ffn_gate_shexp" }, - { LLM_TENSOR_FFN_DOWN_SHEXP, "blk.%d.ffn_down_shexp" }, - { LLM_TENSOR_FFN_UP_SHEXP, "blk.%d.ffn_up_shexp" }, - }, - }, - { - LLM_ARCH_CHATGLM, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - }, - }, - { - LLM_ARCH_BITNET, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_SUB_NORM, "blk.%d.attn_sub_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_SUB_NORM, "blk.%d.ffn_sub_norm" }, - }, - }, - { - LLM_ARCH_T5, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_DEC_OUTPUT_NORM, "dec.output_norm" }, - { LLM_TENSOR_DEC_ATTN_NORM, "dec.blk.%d.attn_norm" }, - { LLM_TENSOR_DEC_ATTN_Q, "dec.blk.%d.attn_q" }, - { LLM_TENSOR_DEC_ATTN_K, "dec.blk.%d.attn_k" }, - { LLM_TENSOR_DEC_ATTN_V, "dec.blk.%d.attn_v" }, - { LLM_TENSOR_DEC_ATTN_OUT, "dec.blk.%d.attn_o" }, - { LLM_TENSOR_DEC_ATTN_REL_B, "dec.blk.%d.attn_rel_b" }, - { LLM_TENSOR_DEC_CROSS_ATTN_NORM, "dec.blk.%d.cross_attn_norm" }, - { LLM_TENSOR_DEC_CROSS_ATTN_Q, "dec.blk.%d.cross_attn_q" }, - { LLM_TENSOR_DEC_CROSS_ATTN_K, "dec.blk.%d.cross_attn_k" }, - { LLM_TENSOR_DEC_CROSS_ATTN_V, "dec.blk.%d.cross_attn_v" }, - { LLM_TENSOR_DEC_CROSS_ATTN_OUT, "dec.blk.%d.cross_attn_o" }, - { LLM_TENSOR_DEC_CROSS_ATTN_REL_B, "dec.blk.%d.cross_attn_rel_b" }, - { LLM_TENSOR_DEC_FFN_NORM, "dec.blk.%d.ffn_norm" }, - { LLM_TENSOR_DEC_FFN_GATE, "dec.blk.%d.ffn_gate" }, - { LLM_TENSOR_DEC_FFN_DOWN, "dec.blk.%d.ffn_down" }, - { LLM_TENSOR_DEC_FFN_UP, "dec.blk.%d.ffn_up" }, - { LLM_TENSOR_ENC_OUTPUT_NORM, "enc.output_norm" }, - { LLM_TENSOR_ENC_ATTN_NORM, "enc.blk.%d.attn_norm" }, - { LLM_TENSOR_ENC_ATTN_Q, "enc.blk.%d.attn_q" }, - { LLM_TENSOR_ENC_ATTN_K, "enc.blk.%d.attn_k" }, - { LLM_TENSOR_ENC_ATTN_V, "enc.blk.%d.attn_v" }, - { LLM_TENSOR_ENC_ATTN_OUT, "enc.blk.%d.attn_o" }, - { LLM_TENSOR_ENC_ATTN_REL_B, "enc.blk.%d.attn_rel_b" }, - { LLM_TENSOR_ENC_FFN_NORM, "enc.blk.%d.ffn_norm" }, - { LLM_TENSOR_ENC_FFN_GATE, "enc.blk.%d.ffn_gate" }, - { LLM_TENSOR_ENC_FFN_DOWN, "enc.blk.%d.ffn_down" }, - { LLM_TENSOR_ENC_FFN_UP, "enc.blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_T5ENCODER, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ENC_OUTPUT_NORM, "enc.output_norm" }, - { LLM_TENSOR_ENC_ATTN_NORM, "enc.blk.%d.attn_norm" }, - { LLM_TENSOR_ENC_ATTN_Q, "enc.blk.%d.attn_q" }, - { LLM_TENSOR_ENC_ATTN_K, "enc.blk.%d.attn_k" }, - { LLM_TENSOR_ENC_ATTN_V, "enc.blk.%d.attn_v" }, - { LLM_TENSOR_ENC_ATTN_OUT, "enc.blk.%d.attn_o" }, - { LLM_TENSOR_ENC_ATTN_REL_B, "enc.blk.%d.attn_rel_b" }, - { LLM_TENSOR_ENC_FFN_NORM, "enc.blk.%d.ffn_norm" }, - { LLM_TENSOR_ENC_FFN_GATE, "enc.blk.%d.ffn_gate" }, - { LLM_TENSOR_ENC_FFN_DOWN, "enc.blk.%d.ffn_down" }, - { LLM_TENSOR_ENC_FFN_UP, "enc.blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_JAIS, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - }, - }, - { - LLM_ARCH_NEMOTRON, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_EXAONE, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ROPE_FREQS, "rope_freqs" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_RWKV6, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_NORM_2, "blk.%d.attn_norm_2" }, - { LLM_TENSOR_TIME_MIX_W1, "blk.%d.time_mix_w1" }, - { LLM_TENSOR_TIME_MIX_W2, "blk.%d.time_mix_w2" }, - { LLM_TENSOR_TIME_MIX_LERP_X, "blk.%d.time_mix_lerp_x" }, - { LLM_TENSOR_TIME_MIX_LERP_W, "blk.%d.time_mix_lerp_w" }, - { LLM_TENSOR_TIME_MIX_LERP_K, "blk.%d.time_mix_lerp_k" }, - { LLM_TENSOR_TIME_MIX_LERP_V, "blk.%d.time_mix_lerp_v" }, - { LLM_TENSOR_TIME_MIX_LERP_R, "blk.%d.time_mix_lerp_r" }, - { LLM_TENSOR_TIME_MIX_LERP_G, "blk.%d.time_mix_lerp_g" }, - { LLM_TENSOR_TIME_MIX_FIRST, "blk.%d.time_mix_first" }, - { LLM_TENSOR_TIME_MIX_DECAY, "blk.%d.time_mix_decay" }, - { LLM_TENSOR_TIME_MIX_DECAY_W1, "blk.%d.time_mix_decay_w1" }, - { LLM_TENSOR_TIME_MIX_DECAY_W2, "blk.%d.time_mix_decay_w2" }, - { LLM_TENSOR_TIME_MIX_KEY, "blk.%d.time_mix_key" }, - { LLM_TENSOR_TIME_MIX_VALUE, "blk.%d.time_mix_value" }, - { LLM_TENSOR_TIME_MIX_RECEPTANCE, "blk.%d.time_mix_receptance" }, - { LLM_TENSOR_TIME_MIX_GATE, "blk.%d.time_mix_gate" }, - { LLM_TENSOR_TIME_MIX_LN, "blk.%d.time_mix_ln" }, - { LLM_TENSOR_TIME_MIX_OUTPUT, "blk.%d.time_mix_output" }, - { LLM_TENSOR_CHANNEL_MIX_LERP_K, "blk.%d.channel_mix_lerp_k" }, - { LLM_TENSOR_CHANNEL_MIX_LERP_R, "blk.%d.channel_mix_lerp_r" }, - { LLM_TENSOR_CHANNEL_MIX_KEY, "blk.%d.channel_mix_key" }, - { LLM_TENSOR_CHANNEL_MIX_VALUE, "blk.%d.channel_mix_value" }, - { LLM_TENSOR_CHANNEL_MIX_RECEPTANCE, "blk.%d.channel_mix_receptance" }, - }, - }, - { - LLM_ARCH_GRANITE, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - }, - }, - { - LLM_ARCH_GRANITE_MOE, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE_INP, "blk.%d.ffn_gate_inp" }, - { LLM_TENSOR_FFN_GATE_EXPS, "blk.%d.ffn_gate_exps" }, - { LLM_TENSOR_FFN_DOWN_EXPS, "blk.%d.ffn_down_exps" }, - { LLM_TENSOR_FFN_UP_EXPS, "blk.%d.ffn_up_exps" }, - }, - }, - { - LLM_ARCH_CHAMELEON, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - { LLM_TENSOR_OUTPUT_NORM, "output_norm" }, - { LLM_TENSOR_OUTPUT, "output" }, - { LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" }, - { LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" }, - { LLM_TENSOR_ATTN_K, "blk.%d.attn_k" }, - { LLM_TENSOR_ATTN_V, "blk.%d.attn_v" }, - { LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" }, - { LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" }, - { LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" }, - { LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" }, - { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, - { LLM_TENSOR_ATTN_Q_NORM, "blk.%d.attn_q_norm" }, - { LLM_TENSOR_ATTN_K_NORM, "blk.%d.attn_k_norm" }, - }, - }, - { - LLM_ARCH_UNKNOWN, - { - { LLM_TENSOR_TOKEN_EMBD, "token_embd" }, - }, - }, -}; - -enum llm_chat_template { - LLM_CHAT_TEMPLATE_CHATML, - LLM_CHAT_TEMPLATE_LLAMA_2, - LLM_CHAT_TEMPLATE_LLAMA_2_SYS, - LLM_CHAT_TEMPLATE_LLAMA_2_SYS_BOS, - LLM_CHAT_TEMPLATE_LLAMA_2_SYS_STRIP, - LLM_CHAT_TEMPLATE_MISTRAL_V1, - LLM_CHAT_TEMPLATE_MISTRAL_V3, - LLM_CHAT_TEMPLATE_MISTRAL_V3_TEKKEN, - LLM_CHAT_TEMPLATE_MISTRAL_V7, - LLM_CHAT_TEMPLATE_PHI_3, - LLM_CHAT_TEMPLATE_FALCON_3, - LLM_CHAT_TEMPLATE_ZEPHYR, - LLM_CHAT_TEMPLATE_MONARCH, - LLM_CHAT_TEMPLATE_GEMMA, - LLM_CHAT_TEMPLATE_ORION, - LLM_CHAT_TEMPLATE_OPENCHAT, - LLM_CHAT_TEMPLATE_VICUNA, - LLM_CHAT_TEMPLATE_VICUNA_ORCA, - LLM_CHAT_TEMPLATE_DEEPSEEK, - LLM_CHAT_TEMPLATE_DEEPSEEK_2, - LLM_CHAT_TEMPLATE_COMMAND_R, - LLM_CHAT_TEMPLATE_LLAMA_3, - LLM_CHAT_TEMPLATE_CHATGML_3, - LLM_CHAT_TEMPLATE_CHATGML_4, - LLM_CHAT_TEMPLATE_MINICPM, - LLM_CHAT_TEMPLATE_EXAONE_3, - LLM_CHAT_TEMPLATE_RWKV_WORLD, - LLM_CHAT_TEMPLATE_GRANITE, - LLM_CHAT_TEMPLATE_GIGACHAT, - LLM_CHAT_TEMPLATE_UNKNOWN, -}; - -static const std::map LLM_CHAT_TEMPLATES = { - { "chatml", LLM_CHAT_TEMPLATE_CHATML }, - { "llama2", LLM_CHAT_TEMPLATE_LLAMA_2 }, - { "llama2-sys", LLM_CHAT_TEMPLATE_LLAMA_2_SYS }, - { "llama2-sys-bos", LLM_CHAT_TEMPLATE_LLAMA_2_SYS_BOS }, - { "llama2-sys-strip", LLM_CHAT_TEMPLATE_LLAMA_2_SYS_STRIP }, - { "mistral-v1", LLM_CHAT_TEMPLATE_MISTRAL_V1 }, - { "mistral-v3", LLM_CHAT_TEMPLATE_MISTRAL_V3 }, - { "mistral-v3-tekken", LLM_CHAT_TEMPLATE_MISTRAL_V3_TEKKEN }, - { "mistral-v7", LLM_CHAT_TEMPLATE_MISTRAL_V7 }, - { "phi3", LLM_CHAT_TEMPLATE_PHI_3 }, - { "falcon3", LLM_CHAT_TEMPLATE_FALCON_3 }, - { "zephyr", LLM_CHAT_TEMPLATE_ZEPHYR }, - { "monarch", LLM_CHAT_TEMPLATE_MONARCH }, - { "gemma", LLM_CHAT_TEMPLATE_GEMMA }, - { "orion", LLM_CHAT_TEMPLATE_ORION }, - { "openchat", LLM_CHAT_TEMPLATE_OPENCHAT }, - { "vicuna", LLM_CHAT_TEMPLATE_VICUNA }, - { "vicuna-orca", LLM_CHAT_TEMPLATE_VICUNA_ORCA }, - { "deepseek", LLM_CHAT_TEMPLATE_DEEPSEEK }, - { "deepseek2", LLM_CHAT_TEMPLATE_DEEPSEEK_2 }, - { "command-r", LLM_CHAT_TEMPLATE_COMMAND_R }, - { "llama3", LLM_CHAT_TEMPLATE_LLAMA_3 }, - { "chatglm3", LLM_CHAT_TEMPLATE_CHATGML_3 }, - { "chatglm4", LLM_CHAT_TEMPLATE_CHATGML_4 }, - { "minicpm", LLM_CHAT_TEMPLATE_MINICPM }, - { "exaone3", LLM_CHAT_TEMPLATE_EXAONE_3 }, - { "rwkv-world", LLM_CHAT_TEMPLATE_RWKV_WORLD }, - { "granite", LLM_CHAT_TEMPLATE_GRANITE }, - { "gigachat", LLM_CHAT_TEMPLATE_GIGACHAT }, -}; - -static llm_arch llm_arch_from_string(const std::string & name) { - for (const auto & kv : LLM_ARCH_NAMES) { // NOLINT - if (kv.second == name) { - return kv.first; - } - } - - return LLM_ARCH_UNKNOWN; + return buft_list; } -// helper to handle gguf constants -// usage: -// -// const auto tn = LLM_TN(LLM_ARCH_LLAMA); -// -// std::string name = tn(LLM_TENSOR_OUTPUT); -> "output" -// std::string name = tn(LLM_TENSOR_TOKEN_EMBD, "bias"); -> "token_embd.bias" -// std::string name = tn(LLM_TENSOR_ATTN_NORM, "weight", 3); -> "blk.3.attn_norm.weight" -// -struct LLM_TN_IMPL { - const llm_arch arch; - const llm_tensor tensor; - const char * const suffix; - const int bid; - const int xid; - - std::string str() const { - if (LLM_TENSOR_NAMES.at(arch).find(tensor) == LLM_TENSOR_NAMES.at(arch).end()) { - return "__missing__"; - } - - std::string name = ::format(LLM_TENSOR_NAMES.at(arch).at(tensor), bid, xid); +// Returns false if cancelled by progress_callback +static bool llm_load_tensors( + llama_model_loader & ml, + llama_model & model, + int n_gpu_layers, + enum llama_split_mode split_mode, + int main_gpu, + const float * tensor_split, + bool use_mlock, + llama_progress_callback progress_callback, + void * progress_callback_user_data) { + auto & hparams = model.hparams; - if (suffix != nullptr) { - name += "."; - name += suffix; - } + model.split_mode = split_mode; + model.main_gpu = main_gpu; + model.n_gpu_layers = n_gpu_layers; - return name; - } + const int n_layer = hparams.n_layer; - operator std::string() const { - return str(); - } + bool use_mmap_buffer = true; - friend bool operator==(const std::string & str, const LLM_TN_IMPL & tn) { - return str == tn.str(); + // build a list of buffer types for the CPU and GPU devices + model.cpu_buft_list = make_cpu_buft_list(model); + for (auto * dev : model.devices) { + llama_model::buft_list_t buft_list = make_gpu_buft_list(dev, split_mode, tensor_split); + // add CPU buffer types as a fallback + buft_list.insert(buft_list.end(), model.cpu_buft_list.begin(), model.cpu_buft_list.end()); + model.gpu_buft_list.emplace(dev, std::move(buft_list)); } - friend bool operator!=(const std::string & str, const LLM_TN_IMPL & tn) { - return str != tn.str(); + // calculate the split points + int device_count = llama_get_device_count(model); + bool all_zero = tensor_split == nullptr || std::all_of(tensor_split, tensor_split + device_count, [](float x) { return x == 0.0f; }); + std::vector splits(device_count); + if (all_zero) { + // default split, by free memory + for (int i = 0; i < device_count; ++i) { + ggml_backend_dev_t dev = model.devices[i]; + size_t total; + size_t free; + ggml_backend_dev_memory(dev, &free, &total); + splits[i] = free; + } + } else { + std::copy(tensor_split, tensor_split + device_count, splits.begin()); } -}; - -struct LLM_TN { - LLM_TN(llm_arch arch) : arch(arch) {} - llm_arch arch; - - LLM_TN_IMPL operator()(llm_tensor tensor, const char * suffix, int bid = -1, int xid = -1) const { - return { arch, tensor, suffix, bid, xid }; + // sum and normalize the splits to get the split points + float split_sum = 0.0f; + for (int i = 0; i < device_count; ++i) { + split_sum += splits[i]; + splits[i] = split_sum; } - - LLM_TN_IMPL operator()(llm_tensor tensor, int bid = -1, int xid = -1) const { - return { arch, tensor, nullptr, bid, xid }; + for (int i = 0; i < device_count; ++i) { + splits[i] /= split_sum; } -}; - -// -// gguf helpers -// - -static const std::map LLAMA_ROPE_SCALING_TYPES = { - { LLAMA_ROPE_SCALING_TYPE_NONE, "none" }, - { LLAMA_ROPE_SCALING_TYPE_LINEAR, "linear" }, - { LLAMA_ROPE_SCALING_TYPE_YARN, "yarn" }, - { LLAMA_ROPE_SCALING_TYPE_LONGROPE, "longrope" }, -}; -static llama_rope_scaling_type llama_rope_scaling_type_from_string(const std::string & name) { - for (const auto & kv : LLAMA_ROPE_SCALING_TYPES) { - if (kv.second == name) { - return (llama_rope_scaling_type) kv.first; + ggml_backend_dev_t cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU); + const int i_gpu_start = std::max((int) hparams.n_layer - n_gpu_layers, (int) 0); + const int act_gpu_layers = model.devices.empty() ? 0 : std::min(n_gpu_layers, (int)n_layer + 1); + auto get_layer_buft_list = [&](int il) -> llama_model::layer_dev { + if (il < i_gpu_start || (il - i_gpu_start) >= act_gpu_layers) { + return {cpu_dev, &model.cpu_buft_list}; } - } + int layer_gpu = std::upper_bound(splits.begin(), splits.begin() + device_count, float(il - i_gpu_start)/act_gpu_layers) - splits.begin(); + auto * dev = model.devices.at(layer_gpu); + return {dev, &model.gpu_buft_list.at(dev)}; + }; - return LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED; -} + // assign the input layer + // there is very little benefit to offloading the input layer, so always keep it on the CPU + model.dev_input = { cpu_dev, &model.cpu_buft_list }; -static std::string gguf_data_to_str(enum gguf_type type, const void * data, int i) { - switch (type) { - case GGUF_TYPE_UINT8: return std::to_string(((const uint8_t *)data)[i]); - case GGUF_TYPE_INT8: return std::to_string(((const int8_t *)data)[i]); - case GGUF_TYPE_UINT16: return std::to_string(((const uint16_t *)data)[i]); - case GGUF_TYPE_INT16: return std::to_string(((const int16_t *)data)[i]); - case GGUF_TYPE_UINT32: return std::to_string(((const uint32_t *)data)[i]); - case GGUF_TYPE_INT32: return std::to_string(((const int32_t *)data)[i]); - case GGUF_TYPE_UINT64: return std::to_string(((const uint64_t *)data)[i]); - case GGUF_TYPE_INT64: return std::to_string(((const int64_t *)data)[i]); - case GGUF_TYPE_FLOAT32: return std::to_string(((const float *)data)[i]); - case GGUF_TYPE_FLOAT64: return std::to_string(((const double *)data)[i]); - case GGUF_TYPE_BOOL: return ((const bool *)data)[i] ? "true" : "false"; - default: return format("unknown type %d", type); + // assign the repeating layers to the devices according to the splits + model.dev_layer.resize(n_layer); + for (int il = 0; il < n_layer; ++il) { + model.dev_layer[il] = get_layer_buft_list(il); } -} + // assign the output layer + model.dev_output = get_layer_buft_list(n_layer); -static std::string gguf_kv_to_str(const struct gguf_context * ctx_gguf, int i) { - const enum gguf_type type = gguf_get_kv_type(ctx_gguf, i); + // one ggml context per buffer type + int max_n_tensors = ml.n_tensors; + max_n_tensors += 1; // duplicated output tensor + max_n_tensors += n_layer*2; // duplicated rope freq tensors + const size_t ctx_size = ggml_tensor_overhead()*max_n_tensors; - switch (type) { - case GGUF_TYPE_STRING: - return gguf_get_val_str(ctx_gguf, i); - case GGUF_TYPE_ARRAY: - { - const enum gguf_type arr_type = gguf_get_arr_type(ctx_gguf, i); - int arr_n = gguf_get_arr_n(ctx_gguf, i); - const void * data = gguf_get_arr_data(ctx_gguf, i); - std::stringstream ss; - ss << "["; - for (int j = 0; j < arr_n; j++) { - if (arr_type == GGUF_TYPE_STRING) { - std::string val = gguf_get_arr_str(ctx_gguf, i, j); - // escape quotes - replace_all(val, "\\", "\\\\"); - replace_all(val, "\"", "\\\""); - ss << '"' << val << '"'; - } else if (arr_type == GGUF_TYPE_ARRAY) { - ss << "???"; - } else { - ss << gguf_data_to_str(arr_type, data, j); - } - if (j < arr_n - 1) { - ss << ", "; - } - } - ss << "]"; - return ss.str(); + std::map ctx_map; + auto ctx_for_buft = [&](ggml_backend_buffer_type_t buft) -> ggml_context * { + auto it = ctx_map.find(buft); + if (it == ctx_map.end()) { + ggml_init_params params = { + /*.mem_size =*/ ctx_size, + /*.mem_buffer =*/ NULL, + /*.no_alloc =*/ true, + }; + ggml_context * ctx = ggml_init(params); + if (!ctx) { + throw std::runtime_error(format("failed to create ggml context")); } - default: - return gguf_data_to_str(type, gguf_get_val_data(ctx_gguf, i), 0); - } -} - -// -// llama helpers -// - -#if defined(_WIN32) -static std::string llama_format_win_err(DWORD err) { - LPSTR buf; - size_t size = FormatMessageA(FORMAT_MESSAGE_ALLOCATE_BUFFER | FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS, - NULL, err, MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), (LPSTR)&buf, 0, NULL); - if (!size) { - return "FormatMessageA failed"; - } - std::string ret(buf, size); - LocalFree(buf); - return ret; -} -#endif - -template -struct no_init { - T value; - no_init() { /* do nothing */ } -}; - -struct llama_file { - -#if defined(_WIN32) - // use FILE * so we don't have to re-open the file to mmap - FILE * fp; - HANDLE fp_win32; - size_t size; - -private: - std::string GetErrorMessageWin32(DWORD error_code) const { - std::string ret; - LPSTR lpMsgBuf = NULL; - DWORD bufLen = FormatMessageA(FORMAT_MESSAGE_ALLOCATE_BUFFER | FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS, - NULL, error_code, MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), (LPSTR)&lpMsgBuf, 0, NULL); - if (!bufLen) { - ret = format("Win32 error code: %lx", error_code); - } else { - ret = lpMsgBuf; - LocalFree(lpMsgBuf); + ctx_map[buft] = ctx; + model.ctxs.emplace_back(ctx); + return ctx; } + return it->second; + }; - return ret; - } - -public: - - llama_file(const char * fname, const char * mode) { - fp = ggml_fopen(fname, mode); - if (fp == NULL) { - throw std::runtime_error(format("failed to open %s: %s", fname, strerror(errno))); - } - fp_win32 = (HANDLE) _get_osfhandle(_fileno(fp)); - seek(0, SEEK_END); - size = tell(); - seek(0, SEEK_SET); - } + // create tensors for the weights + { + // note: cast to int64_t since we will use these for the tensor dimensions + const int64_t n_head = hparams.n_head(); + const int64_t n_head_kv = hparams.n_head_kv(); + const int64_t n_embd = hparams.n_embd; + const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(); + const int64_t n_embd_v_gqa = hparams.n_embd_v_gqa(); + const int64_t n_embd_head_k = hparams.n_embd_head_k; + const int64_t n_embd_head_v = hparams.n_embd_head_v; + const int64_t n_ff = hparams.n_ff(); + const int64_t n_embd_gqa = n_embd_v_gqa; + const int64_t n_vocab = hparams.n_vocab; + const int64_t n_vocab_type = hparams.n_vocab_type; + const int64_t n_rot = hparams.n_rot; + const int64_t n_expert = hparams.n_expert; + const int64_t n_expert_used = hparams.n_expert_used; + const int64_t n_ctx_train = hparams.n_ctx_train; - size_t tell() const { - // SetFilePointerEx returns the current position when seeking relative 0 bytes - LARGE_INTEGER li; - li.QuadPart = 0; - BOOL ret = SetFilePointerEx(fp_win32, li, &li, FILE_CURRENT); - if (!ret) { - throw std::runtime_error(format("read error: %s", GetErrorMessageWin32(GetLastError()).c_str())); + if (n_expert > 0 && hparams.n_expert_used == 0) { + throw std::runtime_error("model has expert layers but no expert layers are used"); } - return li.QuadPart; - } + int n_moved_tensors = 0; + ggml_tensor * first_moved_tensor = nullptr; + ggml_backend_buffer_type_t first_moved_from_buft = nullptr; + ggml_backend_buffer_type_t first_moved_to_buft = nullptr; - void seek(size_t offset, int whence) const { - // no need to convert SEEK_* to FILE_*. The enums are the same. - // Still, keep static asserts to avoid failures in the future. - static_assert(SEEK_SET == FILE_BEGIN, "SEEK_SET != FILE_BEGIN"); - static_assert(SEEK_CUR == FILE_CURRENT, "SEEK_CUR != FILE_CURRENT"); - static_assert(SEEK_END == FILE_END, "SEEK_END != FILE_END"); - - LARGE_INTEGER li; - li.QuadPart = offset; - BOOL ret = SetFilePointerEx(fp_win32, li, NULL, whence); - if (!ret) { - throw std::runtime_error(format("read error: %s", GetErrorMessageWin32(GetLastError()).c_str())); - } - } + auto create_tensor = [&](const LLM_TN_IMPL & tn, const std::initializer_list & ne, int flags) -> ggml_tensor * { + ggml_tensor * t_meta = ml.get_tensor_meta(tn.str().c_str()); - void read_raw(void * ptr, size_t len) const { - // On Win32 ReadFile is significant faster than fread which is again significant faster than std::fstream. Thus - // use the Win32 API to do file io instead of the C/C++ library functions. - - // There are conditions under which ReadFile cannot read chunks >64MB. - // Thus split the operation into smaller chunks if len exceeds this limit. - size_t bytes_read = 0; - while (bytes_read < len) { - size_t chunk_size = std::min(len - bytes_read, 64*1024*1024); - DWORD chunk_read = 0; - BOOL result = ReadFile(fp_win32, reinterpret_cast(ptr) + bytes_read, chunk_size, &chunk_read, NULL); - if (!result) { - throw std::runtime_error(format("read error: %s", GetErrorMessageWin32(GetLastError()).c_str())); - } - if (chunk_read < chunk_size || chunk_read == 0) { - throw std::runtime_error("unexpectedly reached end of file"); + if (!t_meta) { + if (flags & llama_model_loader::TENSOR_NOT_REQUIRED) { + return nullptr; + } + throw std::runtime_error(format("missing tensor '%s'", tn.str().c_str())); } - bytes_read += chunk_read; - } ; - } - - uint32_t read_u32() const { - uint32_t val; - read_raw(&val, sizeof(val)); - return val; - } - - void write_raw(const void * ptr, size_t len) const { - // There are conditions under which WriteFile cannot write chunks >64MB. - // Thus split the operation into smaller chunks if len exceeds this limit. - size_t bytes_written = 0; - while (bytes_written < len) { - size_t chunk_size = std::min(len - bytes_written, 64*1024*1024); - DWORD chunk_written = 0; - BOOL result = WriteFile(fp_win32, reinterpret_cast(ptr) + bytes_written, chunk_size, &chunk_written, NULL); - if (!result) { - throw std::runtime_error(format("write error: %s", GetErrorMessageWin32(GetLastError()).c_str())); + // some models use the token embedding tensor as the output, but since these are used in different layers and with different ops + // the tensor is duplicated + // to handle this, we check if the tensor is duplicated, and if so, we assume that it is being loaded as the output tensor + llm_tensor tn_tensor = tn.tensor; + if (tn.tensor == LLM_TENSOR_TOKEN_EMBD && flags & llama_model_loader::TENSOR_DUPLICATED) { + tn_tensor = LLM_TENSOR_OUTPUT; } - if (chunk_written < chunk_size || chunk_written == 0) { - throw std::runtime_error("unexpectedly failed to write bytes"); + + llm_tensor_info info; + try { + info = llm_tensor_info_for(tn_tensor); + } catch (const std::out_of_range & e) { + throw std::runtime_error(format("missing tensor info mapping for %s", tn.str().c_str())); } - bytes_written += chunk_written; - } - } - - void write_u32(std::uint32_t val) const { - write_raw(&val, sizeof(val)); - } + // tensors with "bias" suffix are always used with GGML_OP_ADD + ggml_op op; + bool bias = tn.suffix != nullptr && strcmp(tn.suffix, "bias") == 0; + if (bias) { + op = GGML_OP_ADD; + } else { + op = info.op; + } - ~llama_file() { - if (fp) { - std::fclose(fp); - } - } -#else - // use FILE * so we don't have to re-open the file to mmap - FILE * fp; - size_t size; - - llama_file(const char * fname, const char * mode) { - fp = ggml_fopen(fname, mode); - if (fp == NULL) { - throw std::runtime_error(format("failed to open %s: %s", fname, strerror(errno))); - } - seek(0, SEEK_END); - size = tell(); - seek(0, SEEK_SET); - } + // sanity checks + if (info.layer == LLM_TENSOR_LAYER_INPUT || info.layer == LLM_TENSOR_LAYER_OUTPUT) { + if (tn.bid != -1) { + GGML_ABORT("input/output layer tensor %s used with a layer number", tn.str().c_str()); + } + } else { + if (tn.bid == -1) { + GGML_ABORT("repeating layer tensor %s used without a layer number", tn.str().c_str()); + } + } - size_t tell() const { -#ifdef _WIN32 - __int64 ret = _ftelli64(fp); -#else - long ret = std::ftell(fp); -#endif - if (ret == -1) { - throw std::runtime_error(format("ftell error: %s", strerror(errno))); - } + // select the buffer type for this tensor + llama_model::buft_list_t * buft_list; + switch (info.layer) { + case LLM_TENSOR_LAYER_INPUT: + buft_list = model.dev_input.buft_list; + break; + case LLM_TENSOR_LAYER_OUTPUT: + buft_list = model.dev_output.buft_list; + break; + case LLM_TENSOR_LAYER_REPEATING: + buft_list = model.dev_layer.at(tn.bid).buft_list; + break; + default: + GGML_ABORT("invalid layer %d for tensor %s", info.layer, tn.str().c_str()); + } - return (size_t) ret; - } + ggml_backend_buffer_type_t buft = select_weight_buft(model, t_meta, op, *buft_list); + if (!buft) { + throw std::runtime_error(format("failed to find a compatible buffer type for tensor %s", tn.str().c_str())); + } - void seek(size_t offset, int whence) const { -#ifdef _WIN32 - int ret = _fseeki64(fp, (__int64) offset, whence); -#else - int ret = std::fseek(fp, (long) offset, whence); -#endif - if (ret != 0) { - throw std::runtime_error(format("seek error: %s", strerror(errno))); - } - } + // avoid using a host buffer when using mmap + auto * buft_dev = ggml_backend_buft_get_device(buft); + if (ml.use_mmap && buft_dev && buft == ggml_backend_dev_host_buffer_type(buft_dev)) { + auto * cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU); + buft = ggml_backend_dev_buffer_type(cpu_dev); + } - void read_raw(void * ptr, size_t len) const { - if (len == 0) { - return; - } - errno = 0; - std::size_t ret = std::fread(ptr, len, 1, fp); - if (ferror(fp)) { - throw std::runtime_error(format("read error: %s", strerror(errno))); - } - if (ret != 1) { - throw std::runtime_error("unexpectedly reached end of file"); - } - } + if (buft != buft_list->front().second) { + n_moved_tensors++; + if (!first_moved_tensor) { + first_moved_tensor = t_meta; + first_moved_from_buft = buft_list->front().second; + first_moved_to_buft = buft; + } + } - uint32_t read_u32() const { - uint32_t ret; - read_raw(&ret, sizeof(ret)); - return ret; - } + ggml_context * ctx = ctx_for_buft(buft); - void write_raw(const void * ptr, size_t len) const { - if (len == 0) { - return; - } - errno = 0; - size_t ret = std::fwrite(ptr, len, 1, fp); - if (ret != 1) { - throw std::runtime_error(format("write error: %s", strerror(errno))); - } - } + // if duplicated, check if the original tensor was allocated in the same buffer type context and avoid creating a new one + if (flags & llama_model_loader::TENSOR_DUPLICATED) { + ggml_tensor * t = ggml_get_tensor(ctx, tn.str().c_str()); + if (t) { + return t; + } + } + return ml.create_tensor(ctx, tn, ne, flags); + }; - void write_u32(std::uint32_t val) const { - write_raw(&val, sizeof(val)); - } + model.layers.resize(n_layer); - ~llama_file() { - if (fp) { - std::fclose(fp); - } - } -#endif -}; -using llama_files = std::vector>; - -struct llama_mmap { - void * addr; - size_t size; - - llama_mmap(const llama_mmap &) = delete; - -#ifdef _POSIX_MAPPED_FILES - static constexpr bool SUPPORTED = true; - - // list of mapped fragments (first_offset, last_offset) - std::vector> mapped_fragments; - - llama_mmap(struct llama_file * file, size_t prefetch = (size_t) -1 /* -1 = max value */, bool numa = false) { - size = file->size; - int fd = fileno(file->fp); - int flags = MAP_SHARED; - // prefetch/readahead impairs performance on NUMA systems - if (numa) { prefetch = 0; } -#ifdef __linux__ - // advise the kernel to read the file sequentially (increases readahead) - if (posix_fadvise(fd, 0, 0, POSIX_FADV_SEQUENTIAL)) { - LLAMA_LOG_WARN("warning: posix_fadvise(.., POSIX_FADV_SEQUENTIAL) failed: %s\n", - strerror(errno)); - } - if (prefetch) { flags |= MAP_POPULATE; } -#endif - addr = mmap(NULL, file->size, PROT_READ, flags, fd, 0); - if (addr == MAP_FAILED) { // NOLINT - throw std::runtime_error(format("mmap failed: %s", strerror(errno))); - } + // TODO: move to a separate function + const auto tn = LLM_TN(model.arch); + switch (model.arch) { + case LLM_ARCH_LLAMA: + case LLM_ARCH_REFACT: + case LLM_ARCH_MINICPM: + case LLM_ARCH_GRANITE: + case LLM_ARCH_GRANITE_MOE: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - if (prefetch > 0) { - // advise the kernel to preload the mapped memory - if (posix_madvise(addr, std::min(file->size, prefetch), POSIX_MADV_WILLNEED)) { - LLAMA_LOG_WARN("warning: posix_madvise(.., POSIX_MADV_WILLNEED) failed: %s\n", - strerror(errno)); - } - } - if (numa) { - // advise the kernel not to use readahead - // (because the next page might not belong on the same node) - if (posix_madvise(addr, file->size, POSIX_MADV_RANDOM)) { - LLAMA_LOG_WARN("warning: posix_madvise(.., POSIX_MADV_RANDOM) failed: %s\n", - strerror(errno)); - } - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); - // initialize list of mapped_fragments - mapped_fragments.emplace_back(0, file->size); - } + // if output is NULL, init from the input tok embed + if (model.output == NULL) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); + } - static void align_range(size_t * first, size_t * last, size_t page_size) { - // align first to the next page - size_t offset_in_page = *first & (page_size - 1); - size_t offset_to_page = offset_in_page == 0 ? 0 : page_size - offset_in_page; - *first += offset_to_page; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - // align last to the previous page - *last = *last & ~(page_size - 1); + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - if (*last <= *first) { - *last = *first; - } - } + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_head_k * n_head}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0); - // partially unmap the file in the range [first, last) - void unmap_fragment(size_t first, size_t last) { - // note: this function must not be called multiple times with overlapping ranges - // otherwise, there is a risk of invalidating addresses that have been repurposed for other mappings - int page_size = sysconf(_SC_PAGESIZE); - align_range(&first, &last, page_size); - size_t len = last - first; + // optional bias tensors + layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - if (len == 0) { - return; - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - GGML_ASSERT(first % page_size == 0); - GGML_ASSERT(last % page_size == 0); - GGML_ASSERT(last > first); + if (hparams.rope_scaling_type_train == LLAMA_ROPE_SCALING_TYPE_LONGROPE) { + layer.rope_long = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG, "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0)); + layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0)); + } + else { + layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0)); + } - void * next_page_start = (uint8_t *) addr + first; + if (n_expert == 0) { + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); - // unmap the range - if (munmap(next_page_start, len)) { - LLAMA_LOG_WARN("warning: munmap failed: %s\n", strerror(errno)); - } + // optional MLP bias + layer.ffn_gate_b = create_tensor(tn(LLM_TENSOR_FFN_GATE, "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED); + } else { + layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0); + layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff, n_expert}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), { n_ff, n_embd, n_expert}, 0); + layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), {n_embd, n_ff, n_expert}, 0); + } + } + } break; + case LLM_ARCH_DECI: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - // update the list of mapped fragments to avoid unmapping the same range again in the destructor - std::vector> new_mapped_fragments; - for (const auto & frag : mapped_fragments) { - if (frag.first < first && frag.second > last) { - // the range is in the middle of the fragment, split it - new_mapped_fragments.emplace_back(frag.first, first); - new_mapped_fragments.emplace_back(last, frag.second); - } else if (frag.first < first && frag.second > first) { - // the range starts in the middle of the fragment - new_mapped_fragments.emplace_back(frag.first, first); - } else if (frag.first < last && frag.second > last) { - // the range ends in the middle of the fragment - new_mapped_fragments.emplace_back(last, frag.second); - } else if (frag.first >= first && frag.second <= last) { - // the range covers the entire fragment - } else { - // the range is outside the fragment - new_mapped_fragments.push_back(frag); - } - } - mapped_fragments = std::move(new_mapped_fragments); - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); - ~llama_mmap() { - for (const auto & frag : mapped_fragments) { - if (munmap((char *) addr + frag.first, frag.second - frag.first)) { - LLAMA_LOG_WARN("warning: munmap failed: %s\n", strerror(errno)); - } - } - } -#elif defined(_WIN32) - static constexpr bool SUPPORTED = true; + // if output is NULL, init from the input tok embed + if (model.output == NULL) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); + } - llama_mmap(struct llama_file * file, size_t prefetch = (size_t) -1, bool numa = false) { - GGML_UNUSED(numa); + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; + const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(i); + const int64_t n_embd_v_gqa = hparams.n_embd_v_gqa(i); + const int64_t n_embd_gqa = hparams.n_embd_v_gqa(i); + const int64_t n_ff = hparams.n_ff(i); + const int64_t n_head = hparams.n_head(i); + const int64_t n_head_kv = hparams.n_head_kv(i); + + if (n_head_kv == 0 && n_head > 0) { + // linear attention for DeciLMCausalModel + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + } + else if (n_head_kv > 0) { + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - size = file->size; + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_head_k * n_head}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0); + } - HANDLE hFile = (HANDLE) _get_osfhandle(_fileno(file->fp)); + // optional bias tensors + layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - HANDLE hMapping = CreateFileMappingA(hFile, NULL, PAGE_READONLY, 0, 0, NULL); + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - if (hMapping == NULL) { - DWORD error = GetLastError(); - throw std::runtime_error(format("CreateFileMappingA failed: %s", llama_format_win_err(error).c_str())); - } + if (hparams.rope_scaling_type_train == LLAMA_ROPE_SCALING_TYPE_LONGROPE) { + layer.rope_long = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG, "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0)); + layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0)); + } + else { + layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0)); + } - addr = MapViewOfFile(hMapping, FILE_MAP_READ, 0, 0, 0); - DWORD error = GetLastError(); - CloseHandle(hMapping); + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); - if (addr == NULL) { - throw std::runtime_error(format("MapViewOfFile failed: %s", llama_format_win_err(error).c_str())); - } + // optional MLP bias + layer.ffn_gate_b = create_tensor(tn(LLM_TENSOR_FFN_GATE, "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED); + } + } break; + case LLM_ARCH_MINICPM3: + { + const int64_t n_embd_head_qk_rope = hparams.n_rot; + const int64_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot; - if (prefetch > 0) { -#if _WIN32_WINNT >= 0x602 - // PrefetchVirtualMemory is only present on Windows 8 and above, so we dynamically load it - BOOL (WINAPI *pPrefetchVirtualMemory) (HANDLE, ULONG_PTR, PWIN32_MEMORY_RANGE_ENTRY, ULONG); - HMODULE hKernel32 = GetModuleHandleW(L"kernel32.dll"); - - // may fail on pre-Windows 8 systems - pPrefetchVirtualMemory = (decltype(pPrefetchVirtualMemory))(void *) GetProcAddress(hKernel32, "PrefetchVirtualMemory"); - - if (pPrefetchVirtualMemory) { - // advise the kernel to preload the mapped memory - WIN32_MEMORY_RANGE_ENTRY range; - range.VirtualAddress = addr; - range.NumberOfBytes = (SIZE_T) std::min(size, prefetch); - if (!pPrefetchVirtualMemory(GetCurrentProcess(), 1, &range, 0)) { - LLAMA_LOG_WARN("warning: PrefetchVirtualMemory failed: %s\n", - llama_format_win_err(GetLastError()).c_str()); - } - } -#else - throw std::runtime_error("PrefetchVirtualMemory unavailable"); -#endif - } - } + const int64_t q_lora_rank = hparams.n_lora_q; + const int64_t kv_lora_rank = hparams.n_lora_kv; + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - void unmap_fragment(size_t first, size_t last) { - // not supported - GGML_UNUSED(first); - GGML_UNUSED(last); - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); - ~llama_mmap() { - if (!UnmapViewOfFile(addr)) { - LLAMA_LOG_WARN("warning: UnmapViewOfFile failed: %s\n", - llama_format_win_err(GetLastError()).c_str()); - } - } -#else - static constexpr bool SUPPORTED = false; + // if output is NULL, init from the input tok embed + if (model.output == NULL) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); + } - llama_mmap(struct llama_file * file, size_t prefetch = -1, bool numa = false) { - GGML_UNUSED(file); - GGML_UNUSED(prefetch); - GGML_UNUSED(numa); + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - throw std::runtime_error("mmap not supported"); - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_q_a_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_A_NORM, "weight", i), {q_lora_rank}, 0); - void unmap_fragment(size_t first, size_t last) { - GGML_UNUSED(first); - GGML_UNUSED(last); + layer.attn_kv_a_norm = create_tensor(tn(LLM_TENSOR_ATTN_KV_A_NORM, "weight", i), {kv_lora_rank}, 0); - throw std::runtime_error("mmap not supported"); - } -#endif -}; -using llama_mmaps = std::vector>; + layer.wq_a = create_tensor(tn(LLM_TENSOR_ATTN_Q_A, "weight", i), {n_embd, q_lora_rank}, 0); + layer.wq_b = create_tensor(tn(LLM_TENSOR_ATTN_Q_B, "weight", i), {q_lora_rank, n_head * n_embd_head_k}, 0); -// Represents some region of memory being locked using mlock or VirtualLock; -// will automatically unlock on destruction. -struct llama_mlock { - void * addr = NULL; - size_t size = 0; + layer.wkv_a_mqa = create_tensor(tn(LLM_TENSOR_ATTN_KV_A_MQA, "weight", i), {n_embd, kv_lora_rank + (n_embd_head_qk_rope)}, 0); + layer.wkv_b = create_tensor(tn(LLM_TENSOR_ATTN_KV_B, "weight", i), {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), { n_head * ( n_embd_head_v), n_embd}, 0); - bool failed_already = false; + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - llama_mlock() {} - llama_mlock(const llama_mlock &) = delete; + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); - ~llama_mlock() { - if (size) { - raw_unlock(addr, size); - } - } + layer.rope_long = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG, "weight", i), { n_embd_head_qk_rope/2 }, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0)); + layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), { n_embd_head_qk_rope/2 }, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0)); + } + } break; + case LLM_ARCH_GROK: + { + if (n_expert == 0) { + throw std::runtime_error("Grok model cannot have zero experts"); + } - void init(void * ptr) { - GGML_ASSERT(addr == NULL && size == 0); // NOLINT - addr = ptr; - } + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - void grow_to(size_t target_size) { - GGML_ASSERT(addr); - if (failed_already) { - return; - } - size_t granularity = lock_granularity(); - target_size = (target_size + granularity - 1) & ~(granularity - 1); - if (target_size > size) { - if (raw_lock((uint8_t *) addr + size, target_size - size)) { - size = target_size; - } else { - failed_already = true; - } - } - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); -#ifdef _POSIX_MEMLOCK_RANGE - static constexpr bool SUPPORTED = true; + // if output is NULL, init from the input tok embed + if (model.output == NULL) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); + } - static size_t lock_granularity() { - return (size_t) sysconf(_SC_PAGESIZE); - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - #ifdef __APPLE__ - #define MLOCK_SUGGESTION \ - "Try increasing the sysctl values 'vm.user_wire_limit' and 'vm.global_user_wire_limit' and/or " \ - "decreasing 'vm.global_no_user_wire_amount'. Also try increasing RLIMIT_MEMLOCK (ulimit -l).\n" - #else - #define MLOCK_SUGGESTION \ - "Try increasing RLIMIT_MEMLOCK ('ulimit -l' as root).\n" - #endif - - bool raw_lock(const void * addr, size_t size) const { - if (!mlock(addr, size)) { - return true; - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - char* errmsg = std::strerror(errno); - bool suggest = (errno == ENOMEM); + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - // Check if the resource limit is fine after all - struct rlimit lock_limit; - if (suggest && getrlimit(RLIMIT_MEMLOCK, &lock_limit)) { - suggest = false; - } - if (suggest && (lock_limit.rlim_max > lock_limit.rlim_cur + size)) { - suggest = false; - } + layer.attn_out_norm = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}, 0); - LLAMA_LOG_WARN("warning: failed to mlock %zu-byte buffer (after previously locking %zu bytes): %s\n%s", - size, this->size, errmsg, suggest ? MLOCK_SUGGESTION : ""); - return false; - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - #undef MLOCK_SUGGESTION + layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0); + layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff, n_expert}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), { n_ff, n_embd, n_expert}, 0); + layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), {n_embd, n_ff, n_expert}, 0); - static void raw_unlock(void * addr, size_t size) { - if (munlock(addr, size)) { - LLAMA_LOG_WARN("warning: failed to munlock buffer: %s\n", std::strerror(errno)); - } - } -#elif defined(_WIN32) - static constexpr bool SUPPORTED = true; + layer.layer_out_norm = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, 0); + } + } break; + case LLM_ARCH_DBRX: + { + if (n_expert == 0) { + throw std::runtime_error("DBRX model cannot have zero experts"); + } - static size_t lock_granularity() { - SYSTEM_INFO si; - GetSystemInfo(&si); - return (size_t) si.dwPageSize; - } + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - bool raw_lock(void * ptr, size_t len) const { - for (int tries = 1; ; tries++) { - if (VirtualLock(ptr, len)) { - return true; - } - if (tries == 2) { - LLAMA_LOG_WARN("warning: failed to VirtualLock %zu-byte buffer (after previously locking %zu bytes): %s\n", - len, size, llama_format_win_err(GetLastError()).c_str()); - return false; - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - // It failed but this was only the first try; increase the working - // set size and try again. - SIZE_T min_ws_size, max_ws_size; - if (!GetProcessWorkingSetSize(GetCurrentProcess(), &min_ws_size, &max_ws_size)) { - LLAMA_LOG_WARN("warning: GetProcessWorkingSetSize failed: %s\n", - llama_format_win_err(GetLastError()).c_str()); - return false; - } - // Per MSDN: "The maximum number of pages that a process can lock - // is equal to the number of pages in its minimum working set minus - // a small overhead." - // Hopefully a megabyte is enough overhead: - size_t increment = len + 1048576; - // The minimum must be <= the maximum, so we need to increase both: - min_ws_size += increment; - max_ws_size += increment; - if (!SetProcessWorkingSetSize(GetCurrentProcess(), min_ws_size, max_ws_size)) { - LLAMA_LOG_WARN("warning: SetProcessWorkingSetSize failed: %s\n", - llama_format_win_err(GetLastError()).c_str()); - return false; - } - } - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - static void raw_unlock(void * ptr, size_t len) { - if (!VirtualUnlock(ptr, len)) { - LLAMA_LOG_WARN("warning: failed to VirtualUnlock buffer: %s\n", - llama_format_win_err(GetLastError()).c_str()); - } - } -#else - static constexpr bool SUPPORTED = false; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - static size_t lock_granularity() { - return (size_t) 65536; - } + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - bool raw_lock(const void * addr, size_t len) const { - LLAMA_LOG_WARN("warning: mlock not supported on this system\n"); - return false; - } + layer.attn_out_norm = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}, 0); - static void raw_unlock(const void * addr, size_t len) {} -#endif -}; -using llama_mlocks = std::vector>; - -// NOTE: avoid ever using this except for building the token_to_piece caches -static std::string llama_token_to_piece(const struct llama_model * model, llama_token token, bool special) { - std::string piece; - piece.resize(piece.capacity()); // using string internal cache - const int n_chars = llama_token_to_piece(model, token, &piece[0], piece.size(), 0, special); - if (n_chars < 0) { - piece.resize(-n_chars); - int check = llama_token_to_piece(model, token, &piece[0], piece.size(), 0, special); - GGML_ASSERT(check == -n_chars); - } - else { - piece.resize(n_chars); - } + layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0); + layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff, n_expert}, 0); + layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff, n_embd, n_expert}, 0); + layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), {n_embd, n_ff, n_expert}, 0); + } + } break; + case LLM_ARCH_BAICHUAN: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); + { + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); + } - return piece; -} + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; -// -// globals -// + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); -struct llama_logger_state { - ggml_log_callback log_callback = llama_log_callback_default; - void * log_callback_user_data = nullptr; -}; + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); -static llama_logger_state g_logger_state; - -// available llama models -enum e_model { - MODEL_UNKNOWN, - MODEL_14M, - MODEL_17M, - MODEL_22M, - MODEL_33M, - MODEL_60M, - MODEL_70M, - MODEL_80M, - MODEL_109M, - MODEL_137M, - MODEL_160M, - MODEL_220M, - MODEL_250M, - MODEL_270M, - MODEL_335M, - MODEL_410M, - MODEL_450M, - MODEL_770M, - MODEL_780M, - MODEL_0_5B, - MODEL_1B, - MODEL_1_3B, - MODEL_1_4B, - MODEL_1_5B, - MODEL_1_6B, - MODEL_2B, - MODEL_2_8B, - MODEL_3B, - MODEL_4B, - MODEL_6B, - MODEL_6_9B, - MODEL_7B, - MODEL_8B, - MODEL_9B, - MODEL_11B, - MODEL_12B, - MODEL_13B, - MODEL_14B, - MODEL_15B, - MODEL_16B, - MODEL_20B, - MODEL_30B, - MODEL_32B, - MODEL_34B, - MODEL_35B, - MODEL_40B, - MODEL_65B, - MODEL_70B, - MODEL_236B, - MODEL_314B, - MODEL_SMALL, - MODEL_MEDIUM, - MODEL_LARGE, - MODEL_XL, - MODEL_A1_7B, - MODEL_A2_7B, - MODEL_8x7B, - MODEL_8x22B, - MODEL_16x12B, - MODEL_10B_128x3_66B, - MODEL_57B_A14B, - MODEL_27B, -}; + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); -static const size_t kiB = 1024; -static const size_t MiB = 1024*kiB; -static const size_t GiB = 1024*MiB; - -struct llama_hparams { - bool vocab_only; - bool rope_finetuned; - bool use_par_res; - bool swin_norm; - - uint32_t n_vocab; - uint32_t n_ctx_train; // context size the model was trained on - uint32_t n_embd; - uint32_t n_layer; - uint32_t n_rot; - uint32_t n_swa = 0; // sliding window attention (SWA) - uint32_t n_embd_head_k; // dimension of keys (d_k). d_q is assumed to be the same, but there are n_head q heads, and only n_head_kv k-v heads - uint32_t n_embd_head_v; // dimension of values (d_v) aka n_embd_head - uint32_t n_expert = 0; - uint32_t n_expert_used = 0; - uint32_t n_vocab_type = 0; // for BERT-style token types - uint32_t n_rel_attn_bkts = 0; - - std::array n_head_arr; - std::array n_head_kv_arr; - std::array n_ff_arr; - - uint32_t n_layer_dense_lead = 0; - uint32_t n_lora_q = 0; - uint32_t n_lora_kv = 0; - uint32_t n_ff_exp = 0; - uint32_t n_ff_shexp = 0; - uint32_t n_expert_shared = 0; - float expert_weights_scale = 0.0; - - float f_norm_eps; - float f_norm_rms_eps; - - float f_attn_logit_softcapping = 50.0f; - float f_final_logit_softcapping = 30.0f; - - // for RWKV - uint32_t rescale_every_n_layers = 0; - uint32_t time_mix_extra_dim = 0; - uint32_t time_decay_extra_dim = 0; - uint32_t wkv_head_size = 0; - - float rope_attn_factor = 1.0f; - float rope_freq_base_train; - float rope_freq_scale_train; - uint32_t n_ctx_orig_yarn; - float rope_yarn_log_mul; - int rope_sections[4]; - - // for State Space Models - uint32_t ssm_d_conv = 0; - uint32_t ssm_d_inner = 0; - uint32_t ssm_d_state = 0; - uint32_t ssm_dt_rank = 0; - bool ssm_dt_b_c_rms = false; - - float f_clamp_kqv = 0.0f; - float f_max_alibi_bias = 0.0f; - float f_logit_scale = 0.0f; - - // Additional scale factors (Granite/Granite MoE) - float f_residual_scale = 0.0f; - float f_embedding_scale = 0.0f; - float f_attention_scale = 0.0f; - - bool causal_attn = true; - bool use_alibi = false; - bool attn_soft_cap = false; - - // needed by encoder-decoder models (e.g. T5, FLAN-T5) - // ref: https://github.com/ggerganov/llama.cpp/pull/8141 - llama_token dec_start_token_id = LLAMA_TOKEN_NULL; - - enum llama_pooling_type pooling_type = LLAMA_POOLING_TYPE_NONE; - enum llama_rope_type rope_type = LLAMA_ROPE_TYPE_NONE; - enum llama_rope_scaling_type rope_scaling_type_train = LLAMA_ROPE_SCALING_TYPE_NONE; - - bool operator!=(const llama_hparams & other) const { - if (this->vocab_only != other.vocab_only) return true; - if (this->n_vocab != other.n_vocab) return true; - if (this->n_ctx_train != other.n_ctx_train) return true; - if (this->n_embd != other.n_embd) return true; - if (this->n_layer != other.n_layer) return true; - if (this->n_rot != other.n_rot) return true; - if (this->n_swa != other.n_swa) return true; - if (this->n_embd_head_k != other.n_embd_head_k) return true; - if (this->n_embd_head_v != other.n_embd_head_v) return true; - if (this->n_expert != other.n_expert) return true; - if (this->n_expert_used != other.n_expert_used) return true; - - if (this->n_head_arr != other.n_head_arr) return true; - if (this->n_head_kv_arr != other.n_head_kv_arr) return true; - if (this->n_ff_arr != other.n_ff_arr) return true; - - if (this->n_rel_attn_bkts != other.n_rel_attn_bkts) return true; - if (this->n_layer_dense_lead != other.n_layer_dense_lead) return true; - if (this->n_lora_q != other.n_lora_q) return true; - if (this->n_lora_kv != other.n_lora_kv) return true; - if (this->n_ff_exp != other.n_ff_exp) return true; - if (this->n_ff_shexp != other.n_ff_shexp) return true; - if (this->n_expert_shared != other.n_expert_shared) return true; - - if (this->rope_finetuned != other.rope_finetuned) return true; - if (this->n_ctx_orig_yarn != other.n_ctx_orig_yarn) return true; - if (std::equal(std::begin(this->rope_sections), - std::end(this->rope_sections), - std::begin(other.rope_sections))) return true; - - if (this->ssm_d_conv != other.ssm_d_conv) return true; - if (this->ssm_d_inner != other.ssm_d_inner) return true; - if (this->ssm_d_state != other.ssm_d_state) return true; - if (this->ssm_dt_rank != other.ssm_dt_rank) return true; - if (this->ssm_dt_b_c_rms != other.ssm_dt_b_c_rms) return true; - - if (this->rescale_every_n_layers != other.rescale_every_n_layers) return true; - if (this->time_mix_extra_dim != other.time_mix_extra_dim) return true; - if (this->time_decay_extra_dim != other.time_decay_extra_dim) return true; - if (this->wkv_head_size != other.wkv_head_size) return true; - - if (this->dec_start_token_id != other.dec_start_token_id) return true; - - const float EPSILON = 1e-9f; - - if (!is_float_close(this->f_norm_eps, other.f_norm_eps, EPSILON)) return true; - if (!is_float_close(this->f_norm_rms_eps, other.f_norm_rms_eps, EPSILON)) return true; - if (!is_float_close(this->rope_attn_factor, other.rope_attn_factor, EPSILON)) return true; - if (!is_float_close(this->rope_freq_base_train, other.rope_freq_base_train, EPSILON)) return true; - if (!is_float_close(this->rope_freq_scale_train, other.rope_freq_scale_train, EPSILON)) return true; - if (!is_float_close(this->expert_weights_scale, other.expert_weights_scale, EPSILON)) return true; - if (!is_float_close(this->rope_yarn_log_mul, other.rope_yarn_log_mul, EPSILON)) return true; - if (!is_float_close(this->f_residual_scale, other.f_residual_scale, EPSILON)) return true; - if (!is_float_close(this->f_embedding_scale, other.f_embedding_scale, EPSILON)) return true; - if (!is_float_close(this->f_attention_scale, other.f_attention_scale, EPSILON)) return true; - - return false; - } + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_FALCON: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - uint32_t n_head(uint32_t il = 0) const { - if (il < n_layer) { - return n_head_arr[il]; - } + // output + { + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); - GGML_ABORT("fatal error"); - } + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); + if (!model.output) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); // needs to be on GPU + } + } - uint32_t n_head_kv(uint32_t il = 0) const { - if (il < n_layer) { - return n_head_kv_arr[il]; - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - GGML_ABORT("fatal error"); - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); - uint32_t n_ff(uint32_t il = 0) const { - if (il < n_layer) { - return n_ff_arr[il]; - } + layer.attn_norm_2 = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.attn_norm_2_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - GGML_ABORT("fatal error"); - } + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - uint32_t n_gqa(uint32_t il = 0) const { - const uint32_t n_head = this->n_head(il); - const uint32_t n_head_kv = this->n_head_kv(il); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_STARCODER: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); + model.pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD, "weight"), {n_embd, n_ctx_train}, 0); - if (n_head_kv == 0) { - return 0; - } + // output + { + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); + if (!model.output) { + // needs to be on GPU + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); + } - return n_head/n_head_kv; - } + } - uint32_t n_embd_k_gqa(uint32_t il = 0) const { // dimension of key embeddings across all k-v heads - const uint32_t n_head_kv = this->n_head_kv(il); + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - return n_embd_head_k * n_head_kv; - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); - uint32_t n_embd_v_gqa(uint32_t il = 0) const { // dimension of value embeddings across all k-v heads - const uint32_t n_head_kv = this->n_head_kv(il); + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0); + layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, 0); - return n_embd_head_v * n_head_kv; - } + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0); - uint32_t n_embd_k_s() const { // dimension of the rolling state embeddings - // corresponds to Mamba's conv_states size or RWKV's token_shift states size - if (wkv_head_size != 0) { - // for RWKV models - return 2 * n_embd; - } else { - // TODO: maybe support other convolution strides than 1 - // NOTE: since the first column of the conv_state is shifted out each time, it's not actually needed - return (ssm_d_conv > 0 ? ssm_d_conv - 1 : 0) * ssm_d_inner; - } - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), {n_embd}, 0); - uint32_t n_embd_v_s() const { // dimension of the recurrent state embeddings - if (wkv_head_size != 0) { - // corresponds to RWKV's wkv_states size - return n_embd * wkv_head_size; - } else { - // corresponds to Mamba's ssm_states size - return ssm_d_state * ssm_d_inner; - } - } -}; + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, 0); -static_assert(std::is_trivially_copyable::value, "llama_hparams must be trivially copyable"); - -struct llama_cparams { - uint32_t n_ctx; // context size used during inference - uint32_t n_batch; - uint32_t n_ubatch; - uint32_t n_seq_max; - int n_threads; // number of threads to use for generation - int n_threads_batch; // number of threads to use for batch processing - - float rope_freq_base; - float rope_freq_scale; - - uint32_t n_ctx_orig_yarn; - // These hyperparameters are not exposed in GGUF, because all - // existing YaRN models use the same values for them. - float yarn_ext_factor; - float yarn_attn_factor; - float yarn_beta_fast; - float yarn_beta_slow; - float defrag_thold; - - bool embeddings; - bool causal_attn; - bool offload_kqv; - bool flash_attn; - bool no_perf; - - enum llama_pooling_type pooling_type; - - ggml_backend_sched_eval_callback cb_eval; - void * cb_eval_user_data; -}; + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, 0); + } + } break; + case LLM_ARCH_BERT: + case LLM_ARCH_NOMIC_BERT: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); + model.type_embd = create_tensor(tn(LLM_TENSOR_TOKEN_TYPES, "weight"), {n_embd, n_vocab_type}, 0); -// TODO: separate into "llama_layer_enc" and "llama_layer_dec" -struct llama_layer { - llama_layer() { - // initialize all pointers to NULL - std::memset(this, 0, sizeof(*this)); - } + if (model.arch == LLM_ARCH_BERT) { + model.pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD, "weight"), {n_embd, n_ctx_train}, 0); - // normalization - struct ggml_tensor * attn_norm; - struct ggml_tensor * attn_norm_b; - struct ggml_tensor * attn_norm_2; - struct ggml_tensor * attn_norm_2_b; - struct ggml_tensor * attn_q_norm; - struct ggml_tensor * attn_q_norm_b; - struct ggml_tensor * attn_k_norm; - struct ggml_tensor * attn_k_norm_b; - struct ggml_tensor * attn_out_norm; - struct ggml_tensor * attn_out_norm_b; - struct ggml_tensor * attn_q_a_norm; - struct ggml_tensor * attn_kv_a_norm; - struct ggml_tensor * attn_sub_norm; - struct ggml_tensor * attn_post_norm; - struct ggml_tensor * ffn_sub_norm; - struct ggml_tensor * attn_norm_cross; - struct ggml_tensor * attn_norm_enc; - - // attention - struct ggml_tensor * wq; - struct ggml_tensor * wk; - struct ggml_tensor * wv; - struct ggml_tensor * wo; - struct ggml_tensor * wqkv; - struct ggml_tensor * wq_a; - struct ggml_tensor * wq_b; - struct ggml_tensor * wkv_a_mqa; - struct ggml_tensor * wkv_b; - struct ggml_tensor * wq_cross; - struct ggml_tensor * wk_cross; - struct ggml_tensor * wv_cross; - struct ggml_tensor * wo_cross; - struct ggml_tensor * wq_enc; - struct ggml_tensor * wk_enc; - struct ggml_tensor * wv_enc; - struct ggml_tensor * wo_enc; - - // attention bias - struct ggml_tensor * bq; - struct ggml_tensor * bk; - struct ggml_tensor * bv; - struct ggml_tensor * bo; - struct ggml_tensor * bqkv; - - // relative position bias - struct ggml_tensor * attn_rel_b; - struct ggml_tensor * attn_rel_b_enc; - struct ggml_tensor * attn_rel_b_cross; - - // normalization - struct ggml_tensor * ffn_norm; - struct ggml_tensor * ffn_norm_b; - struct ggml_tensor * ffn_post_norm; - struct ggml_tensor * layer_out_norm; - struct ggml_tensor * layer_out_norm_b; - struct ggml_tensor * ffn_norm_exps; - struct ggml_tensor * ffn_norm_enc; - - // ff - struct ggml_tensor * ffn_gate; // w1 - struct ggml_tensor * ffn_down; // w2 - struct ggml_tensor * ffn_up; // w3 - struct ggml_tensor * ffn_gate_enc; - struct ggml_tensor * ffn_down_enc; - struct ggml_tensor * ffn_up_enc; - - // ff MoE - struct ggml_tensor * ffn_gate_inp; - struct ggml_tensor * ffn_gate_exps; - struct ggml_tensor * ffn_down_exps; - struct ggml_tensor * ffn_up_exps ; - - // ff shared expert (shexp) - struct ggml_tensor * ffn_gate_inp_shexp; - struct ggml_tensor * ffn_gate_shexp; - struct ggml_tensor * ffn_down_shexp; - struct ggml_tensor * ffn_up_shexp; - - // ff bias - struct ggml_tensor * ffn_gate_b; - struct ggml_tensor * ffn_down_b; // b2 - struct ggml_tensor * ffn_up_b; // b3 - struct ggml_tensor * ffn_act; - - // mamba proj - struct ggml_tensor * ssm_in; - struct ggml_tensor * ssm_x; - struct ggml_tensor * ssm_dt; - struct ggml_tensor * ssm_out; - - // mamba - struct ggml_tensor * ssm_conv1d; - struct ggml_tensor * ssm_a; - struct ggml_tensor * ssm_d; - - // mamba bias - struct ggml_tensor * ssm_conv1d_b; - struct ggml_tensor * ssm_dt_b; - - // rwkv - struct ggml_tensor * time_mix_w1; - struct ggml_tensor * time_mix_w2; - struct ggml_tensor * time_mix_lerp_x; - struct ggml_tensor * time_mix_lerp_w; - struct ggml_tensor * time_mix_lerp_k; - struct ggml_tensor * time_mix_lerp_v; - struct ggml_tensor * time_mix_lerp_r; - struct ggml_tensor * time_mix_lerp_g; - - struct ggml_tensor * time_mix_first; - struct ggml_tensor * time_mix_decay; - struct ggml_tensor * time_mix_decay_w1; - struct ggml_tensor * time_mix_decay_w2; - struct ggml_tensor * time_mix_key; - struct ggml_tensor * time_mix_value; - struct ggml_tensor * time_mix_receptance; - struct ggml_tensor * time_mix_gate; - - struct ggml_tensor * time_mix_ln; - struct ggml_tensor * time_mix_ln_b; - struct ggml_tensor * time_mix_output; - - struct ggml_tensor * channel_mix_lerp_k; - struct ggml_tensor * channel_mix_lerp_r; - - struct ggml_tensor * channel_mix_key; - struct ggml_tensor * channel_mix_receptance; - struct ggml_tensor * channel_mix_value; - - // long rope factors - struct ggml_tensor * rope_long = nullptr; - struct ggml_tensor * rope_short = nullptr; - struct ggml_tensor * rope_freqs = nullptr; - - // bitnet scale - struct ggml_tensor * wq_scale; - struct ggml_tensor * wk_scale; - struct ggml_tensor * wv_scale; - struct ggml_tensor * wo_scale; - struct ggml_tensor * ffn_gate_scale; - struct ggml_tensor * ffn_up_scale; - struct ggml_tensor * ffn_down_scale; -}; + model.cls = create_tensor(tn(LLM_TENSOR_CLS, "weight"), {n_embd, n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + model.cls_b = create_tensor(tn(LLM_TENSOR_CLS, "bias"), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); -// very similar to llama_batch, -// but has more metadata about sequences -struct llama_ubatch { - bool equal_seqs; - // TODO: whole_seqs for embeddings? - - uint32_t n_tokens; // total tokens (n_seq_tokens * n_seqs) - uint32_t n_seq_tokens; // tokens per sequence - uint32_t n_seqs; - - llama_token * token; // [n_tokens] - float * embd; // [n_embd, n_tokens] - llama_pos * pos; // [n_tokens] - int32_t * n_seq_id; // [n_seqs] - llama_seq_id ** seq_id; // [n_seqs] - int8_t * output; // [n_tokens] -}; + model.cls_out = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), {n_embd, 1}, llama_model_loader::TENSOR_NOT_REQUIRED); + model.cls_out_b = create_tensor(tn(LLM_TENSOR_CLS_OUT, "bias"), {1}, llama_model_loader::TENSOR_NOT_REQUIRED); + } -struct llama_kv_cell { - llama_pos pos = -1; - llama_pos delta = 0; - int32_t src = -1; // used by recurrent state models to copy states - int32_t tail = -1; + model.tok_norm = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight"), {n_embd}, 0); + model.tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias"), {n_embd}, 0); - std::set seq_id; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - bool has_seq_id(const llama_seq_id & id) const { - return seq_id.find(id) != seq_id.end(); - } + if (model.arch == LLM_ARCH_BERT) { + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, 0); - bool is_empty() const { - return seq_id.empty(); - } + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, 0); - bool is_same_seq(const llama_kv_cell & other) const { - return seq_id == other.seq_id; - } -}; + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, 0); + } else { + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0); + } -// ring-buffer of cached KV data -struct llama_kv_cache { - bool has_shift = false; - bool do_defrag = false; - bool recurrent = false; // with recurrent state models, a cell can hold the state for more than one past token - bool v_trans = true; // the value tensor is transposed + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - // Note: The value of head isn't only used to optimize searching - // for a free KV slot. llama_decode_internal also uses it, so it - // cannot be freely changed after a slot has been allocated. - uint32_t head = 0; - uint32_t size = 0; - uint32_t used = 0; // used cells (i.e. at least one seq_id) + layer.attn_out_norm = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}, 0); + layer.attn_out_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "bias", i), {n_embd}, 0); - // computed before each graph build - uint32_t n = 0; + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); - ggml_type type_k = GGML_TYPE_F16; - ggml_type type_v = GGML_TYPE_F16; + if (model.arch == LLM_ARCH_BERT) { + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, 0); + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, 0); + } else { + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + } - std::vector cells; + layer.layer_out_norm = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, 0); + layer.layer_out_norm_b = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "bias", i), {n_embd}, 0); + } + } break; + case LLM_ARCH_JINA_BERT_V2: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); // word_embeddings + model.type_embd = create_tensor(tn(LLM_TENSOR_TOKEN_TYPES, "weight"), {n_embd, n_vocab_type}, 0); // token_type_embeddings - std::vector k_l; // per layer - std::vector v_l; + model.tok_norm = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight"), {n_embd}, 0); // LayerNorm + model.tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias"), {n_embd}, 0); //LayerNorm bias - std::vector ctxs; - std::vector bufs; + model.cls = create_tensor(tn(LLM_TENSOR_CLS, "weight"), {n_embd, 1}, llama_model_loader::TENSOR_NOT_REQUIRED); + model.cls_b = create_tensor(tn(LLM_TENSOR_CLS, "bias"), {1}, llama_model_loader::TENSOR_NOT_REQUIRED); + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; // JinaBertLayer - size_t total_size() { - size_t size = 0; - for (auto & buf : bufs) { - size += ggml_backend_buffer_get_size(buf.get()); - } - return size; - } -}; + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, 0); -struct llama_control_vector { - std::vector tensors; // per layer - std::vector ctxs; - std::vector bufs; + layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.attn_q_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - int32_t layer_start = -1; - int32_t layer_end = -1; + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, 0); - struct ggml_tensor * tensor_for(int il) const { - if (il < 0 || il < layer_start || il > layer_end || (size_t) il >= tensors.size()) { - return nullptr; - } - return tensors[il]; - } + layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.attn_k_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - struct ggml_tensor * apply_to(struct ggml_context * ctx, struct ggml_tensor * cur, int il) const { - ggml_tensor * layer_dir = tensor_for(il); - if (layer_dir != nullptr) { - cur = ggml_add(ctx, cur, layer_dir); - } - return cur; - } -}; + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, 0); -struct llama_model { - e_model type = MODEL_UNKNOWN; - llm_arch arch = LLM_ARCH_UNKNOWN; - llama_ftype ftype = LLAMA_FTYPE_ALL_F32; + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); //output_dens + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0); //output_dens - std::string name = "n/a"; + layer.attn_out_norm = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}, 0); //output_norm + layer.attn_out_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "bias", i), {n_embd}, 0); - llama_hparams hparams = {}; - llama_vocab vocab; + layer.attn_norm_2 = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.attn_norm_2_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - struct ggml_tensor * tok_embd = nullptr; - struct ggml_tensor * type_embd = nullptr; - struct ggml_tensor * pos_embd = nullptr; - struct ggml_tensor * tok_norm = nullptr; - struct ggml_tensor * tok_norm_b = nullptr; - - struct ggml_tensor * output_norm = nullptr; - struct ggml_tensor * output_norm_b = nullptr; - struct ggml_tensor * output = nullptr; - struct ggml_tensor * output_b = nullptr; - struct ggml_tensor * output_norm_enc = nullptr; + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); - // classifier - struct ggml_tensor * cls = nullptr; - struct ggml_tensor * cls_b = nullptr; - struct ggml_tensor * cls_out = nullptr; - struct ggml_tensor * cls_out_b = nullptr; + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, 0); - std::vector layers; + layer.layer_out_norm = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, 0); + layer.layer_out_norm_b = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "bias", i), {n_embd}, 0); + } + } break; + case LLM_ARCH_BLOOM: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); + model.tok_norm = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight"), {n_embd}, 0); + model.tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias"), {n_embd}, 0); - // gguf metadata - std::unordered_map gguf_kv; + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - llama_split_mode split_mode; - int main_gpu; - int n_gpu_layers; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - std::vector rpc_servers; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); - // list of devices used in this model - std::vector devices; + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0); + layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0); - // lists of buffer types used for each layer - using buft_list_t = std::vector>; - buft_list_t cpu_buft_list; - std::map gpu_buft_list; + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), {n_embd}, 0); - struct layer_dev { - ggml_backend_dev_t dev; - buft_list_t * buft_list; - }; - layer_dev dev_input = {}; - layer_dev dev_output = {}; - std::vector dev_layer; + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, 0); - // contexts where the model tensors metadata is stored - std::vector ctxs; + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, 0); + } + } break; + case LLM_ARCH_MPT: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); + model.pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD, "weight"), {n_embd, n_ctx_train}, llama_model_loader::TENSOR_NOT_REQUIRED); - // the model memory buffers for the tensor data - std::vector bufs; + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - // model memory mapped files - llama_mmaps mappings; + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); + if (!model.output) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); // needs to be on GPU + } - // objects representing data potentially being locked in memory - llama_mlocks mlock_bufs; - llama_mlocks mlock_mmaps; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - // for quantize-stats only - std::vector> tensors_by_name; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - int64_t t_load_us = 0; - int64_t t_start_us = 0; + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0); + layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); - // total number of parameters in the model - uint64_t n_elements = 0; + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - // total size of all the tensors in the model in bytes - size_t n_bytes = 0; + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - // keep track of loaded lora adapters - std::set lora_adapters; + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - ~llama_model() { - while (!lora_adapters.empty()) { - llama_lora_adapter_free(*lora_adapters.begin()); - } - } -}; + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED); -struct llama_sbatch_seq { - int32_t n_seq_id; - llama_seq_id * seq_id; - size_t offset; - size_t length; -}; + layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.attn_q_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); -// sequence-length-aware batch splitting -struct llama_sbatch { - // tokens left in this batch - size_t n_tokens; - - size_t n_embd; - - bool logits_all; // TODO: remove once lctx.logits_all is removed too - - // sorted indices into the batch - std::vector ids; - // batch indices of the output - std::vector out_ids; - std::vector seq; - const llama_batch * batch = nullptr; - - // buffers for the ubatch - std::vector ubatch_token; - std::vector ubatch_embd; - std::vector ubatch_pos; - std::vector ubatch_n_seq_id; - std::vector ubatch_seq_id; - std::vector ubatch_output; - - llama_ubatch reserve_ubatch(size_t n_ubatch, bool has_embd = false) { - // clear empty sequences - // the previous ubatch is assumed to be gone, - // so nothing should refer to values in these sequences anymore. - for (size_t i = seq.size(); i-- > 0;) { - if (seq[i].length == 0) { - seq.pop_back(); - } else { - break; - } - } - ubatch_token.resize(!has_embd ? n_ubatch : 0); - ubatch_embd.resize(has_embd ? n_embd * n_ubatch : 0); - ubatch_pos.resize(n_ubatch); - ubatch_n_seq_id.resize(n_ubatch); - ubatch_seq_id.resize(n_ubatch); - ubatch_output.resize(n_ubatch); - llama_ubatch ubatch = { - /*equal_seqs =*/ true, - /*n_tokens =*/ 0, - /*n_seq_tokens =*/ 0, - /*n_seqs =*/ 0, - /*token =*/ !has_embd ? ubatch_token.data() : nullptr, - /*embd =*/ has_embd ? ubatch_embd.data() : nullptr, - /*pos =*/ ubatch_pos.data(), - /*n_seq_id =*/ ubatch_n_seq_id.data(), - /*seq_id =*/ ubatch_seq_id.data(), - /*output =*/ ubatch_output.data(), - }; - return ubatch; - } + layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.attn_k_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - void add_seq_to_ubatch(llama_ubatch & ubatch, llama_sbatch_seq & seq, size_t length) { - GGML_ASSERT(batch != nullptr); - GGML_ASSERT(length <= seq.length); - // Can only add sequences of equal lengths to a batch, - // otherwise it isn't clear to which sequence a token belongs - GGML_ASSERT(seq.n_seq_id == 0 || ubatch.n_seqs == 0 || length == (size_t) ubatch.n_tokens / ubatch.n_seqs); - GGML_ASSERT((seq.n_seq_id != 0) == ubatch.equal_seqs); - // NOTE: loops are separated for cache-friendliness - if (batch->token) { - if (ubatch.equal_seqs) { - for (size_t i = 0; i < length; ++i) { - ubatch.token[ubatch.n_tokens + i] = batch->token[ids[seq.offset + i]]; - } - } else { - // simple split - ubatch.token = batch->token + seq.offset; - } - } else { - ubatch.token = nullptr; - } - if (batch->embd) { - if (ubatch.equal_seqs) { - for (size_t i = 0; i < length; ++i) { - memcpy( - ubatch.embd + n_embd * (ubatch.n_tokens + i), - batch->embd + n_embd * ids[seq.offset + i], - n_embd * sizeof(float) - ); - } - } else { - // simple split - ubatch.embd = batch->embd + (n_embd * seq.offset); - } - } else { - ubatch.embd = nullptr; - } - if (ubatch.equal_seqs) { - for (size_t i = 0; i < length; ++i) { - ubatch.pos[ubatch.n_tokens + i] = batch->pos[ids[seq.offset + i]]; - } - } else { - // simple split - ubatch.pos = batch->pos + seq.offset; - } - if (ubatch.equal_seqs) { - ubatch.n_seq_id[ubatch.n_seqs] = seq.n_seq_id; - if (seq.seq_id) { - ubatch.seq_id[ubatch.n_seqs] = seq.seq_id; - } - } else { - // simple split - if (batch->n_seq_id) { - ubatch.n_seq_id = batch->n_seq_id + seq.offset; - } else { - for (size_t i = 0; i < length; ++i) { - ubatch.n_seq_id[ubatch.n_seqs + i] = 1; - } - } - if (batch->seq_id) { - ubatch.seq_id = batch->seq_id + seq.offset; - } - } - if (logits_all) { - for (size_t i = 0; i < length; ++i) { - ubatch.output[ubatch.n_tokens + i] = 1; - out_ids.push_back(ids[seq.offset + i]); - } - } else if (batch->logits) { - if (ubatch.equal_seqs) { - for (size_t i = 0; i < length; ++i) { - size_t id = ids[seq.offset + i]; - int8_t is_output = batch->logits[id]; - ubatch.output[ubatch.n_tokens + i] = is_output; - if (is_output) { out_ids.push_back(id); } - } - } else { - // simple split - ubatch.output = batch->logits + seq.offset; - for (size_t i = 0; i < length; ++i) { - if (ubatch.output[i] != 0) { out_ids.push_back(seq.offset + i); } - } - } - } else { - // only get last output - for (size_t i = 0; i < length; ++i) { - size_t id = ids[seq.offset + i]; - int8_t is_last = id == ids.size() - 1; - ubatch.output[ubatch.n_tokens + i] = is_last; - if (is_last) { out_ids.push_back(id); } - } - } - if (ubatch.n_tokens == 0 && ubatch.n_seqs == 0) { - ubatch.n_seq_tokens = ubatch.equal_seqs ? length : 1; - } - ubatch.n_tokens += length; - ubatch.n_seqs += ubatch.equal_seqs ? 1 : length; // virtual sequences for simple splits - seq.offset += length; - seq.length -= length; - n_tokens -= length; - GGML_ASSERT(ubatch.n_tokens == ubatch.n_seq_tokens * ubatch.n_seqs); - } + // AWQ ScaleActivation layer + layer.ffn_act = create_tensor(tn(LLM_TENSOR_FFN_ACT, "scales", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED); + } + } break; + case LLM_ARCH_STABLELM: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - // simple split, unknown number of sequences of unequal lengths - llama_ubatch split_simple(size_t n_ubatch) { - n_ubatch = n_tokens < n_ubatch ? n_tokens : n_ubatch; - llama_ubatch ubatch = reserve_ubatch(n_ubatch, /* has_embd */ batch->embd != nullptr); - ubatch.equal_seqs = false; - if (!seq.empty()) { - llama_sbatch_seq & s = seq[0]; - size_t length = s.length < n_ubatch ? s.length : n_ubatch; - GGML_ASSERT(seq.size() == 1 && s.n_seq_id == 0); // don't mix with other splits - add_seq_to_ubatch(ubatch, s, length); - } - return ubatch; - } + // output + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - // make batches of equal-length sequences - llama_ubatch split_equal(size_t n_ubatch) { - n_ubatch = n_tokens < n_ubatch ? n_tokens : n_ubatch; - llama_ubatch ubatch = reserve_ubatch(n_ubatch, /* has_embd */ batch->embd != nullptr); - if (!seq.empty()) { - size_t length = 0; - size_t n_tokens_in_ubatch = 0; - GGML_ASSERT(seq[0].n_seq_id > 0); // should not be mixed with simple splits - // smallest first, because it's easier to split this way; - // starting from the end to pop in constant time. - for (size_t i = seq.size(); i-- > 0;) { - llama_sbatch_seq & s = seq[i]; - GGML_ASSERT(s.length > 0); - if (length == 0) { - length = s.length < n_ubatch ? s.length : n_ubatch; - } - add_seq_to_ubatch(ubatch, s, length); - n_tokens_in_ubatch += length; - // shared prompts can't be mixed with any of their sequences, - // so it's safer to compute them in their own ubatch - if (s.n_seq_id > 1) { break; } - // stop when there isn't enough space for another sequence - if (length + n_tokens_in_ubatch > n_ubatch) { break; } - } - } - return ubatch; - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - // sequence-wise split - llama_ubatch split_seq(size_t n_ubatch) { - n_ubatch = n_tokens < n_ubatch ? n_tokens : n_ubatch; - llama_ubatch ubatch = reserve_ubatch(n_ubatch, /* has_embd */ batch->embd != nullptr); - if (!seq.empty()) { - llama_sbatch_seq & s = seq[seq.size() - 1]; - size_t length = s.length < n_ubatch ? s.length : n_ubatch; - GGML_ASSERT(s.n_seq_id > 0); // should not be mixed with simple splits - add_seq_to_ubatch(ubatch, s, length); - } - return ubatch; - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); - void from_batch(const llama_batch & batch, const size_t n_embd, const bool simple_split = false, const bool logits_all = false) { - GGML_ASSERT(batch.n_tokens >= 0); - this->batch = &batch; - this->n_embd = n_embd; - this->logits_all = logits_all; + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - n_tokens = batch.n_tokens; - ids.resize(n_tokens); - out_ids.clear(); - // TODO: reserve out_ids and seq + // optional bias tensors, present in Stable LM 2 1.6B + layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); - for (size_t i = 0; i < n_tokens; ++i) { - ids[i] = i; - } - if (simple_split) { - seq.resize(1); - llama_sbatch_seq & s = seq[0]; - s.n_seq_id = 0; - s.seq_id = nullptr; - s.offset = 0; - s.length = n_tokens; - return; - } - std::sort(ids.begin(), ids.end(), - [&batch](size_t a, size_t b) { - int32_t n_seq_a = batch.n_seq_id ? batch.n_seq_id[a] : 1; - int32_t n_seq_b = batch.n_seq_id ? batch.n_seq_id[b] : 1; - // sort by seq_id, then by pos - if (n_seq_a == n_seq_b) { - if (batch.seq_id) { - for (int32_t i = 0; i < n_seq_a; ++i) { - llama_seq_id seq_id_a = batch.seq_id[a][i]; - llama_seq_id seq_id_b = batch.seq_id[b][i]; - // smaller seq_ids go first - if (seq_id_a != seq_id_b) { - return seq_id_a < seq_id_b; - } - } - } - // when all else is equal, sort by pos - if (batch.pos) { - return batch.pos[a] < batch.pos[b]; - } - // no pos, sort by id - return a < b; - } - // shared prompts go first - return n_seq_a > n_seq_b; - } - ); - // init seq - llama_sbatch_seq * last_seq = nullptr; - - for (size_t i = 0; i < n_tokens; ++i) { - const size_t bi = ids[i]; - const int32_t n_seqs = batch.n_seq_id[bi]; - llama_seq_id * seq_ids = batch.seq_id[bi]; - if (last_seq != nullptr) { - bool same = n_seqs == last_seq->n_seq_id; - for (int32_t j = 0; same && j < n_seqs; ++j) { - if (seq_ids[j] != last_seq->seq_id[j]) { - same = false; - } - } - if (same) { - last_seq->length += 1; - continue; - } - } - llama_sbatch_seq new_seq = {n_seqs, seq_ids, i, 1}; - seq.push_back(new_seq); - last_seq = &seq.back(); - } - // keep shared prompts first at the end, then sort by length descending. - std::sort(seq.begin(), seq.end(), - [](llama_sbatch_seq & a, llama_sbatch_seq & b) { - if (a.n_seq_id == b.n_seq_id) { - return a.length > b.length; - } - return a.n_seq_id < b.n_seq_id; - } - ); - } -}; + // optional q and k layernorms, present in StableLM 2 12B + layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k, n_head}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k, n_head_kv}, llama_model_loader::TENSOR_NOT_REQUIRED); -struct llama_context { - llama_context(const llama_model & model) - : model(model) - , t_start_us(model.t_start_us) - , t_load_us(model.t_load_us) {} + // optional FFN norm, not present in StableLM 2 12B which uses parallel residual + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - const struct llama_model & model; + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_QWEN: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - struct llama_cparams cparams; - struct llama_sbatch sbatch; - struct llama_kv_cache kv_self; - struct llama_control_vector cvec; + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - std::unordered_map lora_adapters; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - std::vector backends; - std::vector> set_n_threads_fns; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - ggml_backend_t backend_cpu = nullptr; + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd*3}, 0); + layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd*3}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - ggml_threadpool_t threadpool = nullptr; - ggml_threadpool_t threadpool_batch = nullptr; + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - bool has_evaluated_once = false; + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff/2}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff/2, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff/2}, 0); + } + } break; + case LLM_ARCH_QWEN2: + case LLM_ARCH_QWEN2VL: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - mutable int64_t t_start_us; - mutable int64_t t_load_us; - mutable int64_t t_p_eval_us = 0; - mutable int64_t t_eval_us = 0; + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); + // if output is NULL, init from the input tok embed + if (model.output == NULL) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); + } - mutable int64_t t_compute_start_us = 0; - mutable int64_t n_queued_tokens = 0; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - mutable int32_t n_p_eval = 0; // number of tokens in eval calls for the prompt (with batch size > 1) - mutable int32_t n_eval = 0; // number of eval calls + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - // host buffer for the model output (logits and embeddings) - ggml_backend_buffer_ptr buf_output; + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - // decode output (2-dimensional array: [n_outputs][n_vocab]) - size_t logits_size = 0; // capacity (of floats) for logits - float * logits = nullptr; + // optional bias tensors + layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, 0); + layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, 0); + layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, 0); - std::vector output_ids; // map batch token positions to ids of the logits and embd buffers - size_t output_size = 0; // capacity (of tokens positions) for the output buffers - int32_t n_outputs = 0; // number of actually-used outputs in the current ubatch or last logical batch + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - bool logits_all = false; + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_QWEN2MOE: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - // embeddings output (2-dimensional array: [n_outputs][n_embd]) - // populated only when pooling_type == LLAMA_POOLING_TYPE_NONE - size_t embd_size = 0; // capacity (of floats) for embeddings - float * embd = nullptr; + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - // sequence embeddings output (map of [n_embd] vectors) - // populated only when pooling_type != LLAMA_POOLING_TYPE_NONE - std::map> embd_seq; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - // whether we are computing encoder output or decoder output - bool is_encoding = false; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - // TODO: find a better way to accommodate mutli-dimension position encoding methods - // number of position id each token get, 1 for each token in most cases. - // when using m-rope, it will be 3 position ids per token to representing 3 dimension coordinate. - int n_pos_per_token = 1; + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - // output of the encoder part of the encoder-decoder models - std::vector embd_enc; - std::vector> seq_ids_enc; + // optional bias tensors + layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, 0); + layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, 0); + layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, 0); - // memory buffers used to evaluate the model - std::vector buf_compute_meta; - ggml_backend_sched_ptr sched; + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - ggml_abort_callback abort_callback = nullptr; - void * abort_callback_data = nullptr; + layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0); - // input tensors - struct ggml_tensor * inp_tokens; // I32 [n_batch] - struct ggml_tensor * inp_embd; // F32 [n_embd, n_batch] - struct ggml_tensor * inp_pos; // I32 [n_batch] - struct ggml_tensor * inp_out_ids; // I32 [n_outputs] - struct ggml_tensor * inp_KQ_mask; // F32 [kv_size, n_batch] - struct ggml_tensor * inp_KQ_mask_swa; // F32 [kv_size, n_batch] - struct ggml_tensor * inp_K_shift; // I32 [kv_size] - struct ggml_tensor * inp_mean; // F32 [n_batch, n_batch] - struct ggml_tensor * inp_cls; // I32 [n_batch] - struct ggml_tensor * inp_s_copy; // I32 [kv_size] - struct ggml_tensor * inp_s_mask; // F32 [1, n_kv] - struct ggml_tensor * inp_s_seq; // I32 [n_kv, n_batch] - struct ggml_tensor * inp_pos_bucket; // I32 [n_batch|n_kv, n_batch] - struct ggml_tensor * inp_embd_enc; // F32 [n_embd, n_outputs_enc] - struct ggml_tensor * inp_KQ_mask_cross; // F32 [n_outputs_enc, n_batch] -}; + if (n_expert == 0) { + throw std::runtime_error("n_expert must be > 0 for QWEN2MOE"); + } + if (n_expert_used == 0) { + throw std::runtime_error("n_expert_used must be > 0 for QWEN2MOE"); + } -struct llama_lora_weight { - struct ggml_tensor * a = nullptr; - struct ggml_tensor * b = nullptr; - llama_lora_weight() = default; - llama_lora_weight(struct ggml_tensor * a, struct ggml_tensor * b): a(a), b(b) {} -}; + // MoE branch + const int64_t n_ff_exp = hparams.n_ff_exp ? hparams.n_ff_exp : n_ff / n_expert_used; -struct llama_lora_adapter { - struct llama_model * base_model; - // map tensor name to lora_a_b - std::unordered_map ab_map; - std::vector ctxs; - std::vector bufs; + layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), { n_embd, n_ff_exp, n_expert}, 0); + layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp, n_embd, n_expert}, 0); + layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), { n_embd, n_ff_exp, n_expert}, 0); - float alpha; + // Shared expert branch + const int64_t n_ff_shexp = hparams.n_ff_shexp ? hparams.n_ff_shexp : n_ff; - llama_lora_adapter(struct llama_model * base_model): base_model(base_model) { - base_model->lora_adapters.insert(this); - } + layer.ffn_gate_inp_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP_SHEXP, "weight", i), {n_embd}, 0); + layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), { n_embd, n_ff_shexp}, 0); + layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), {n_ff_shexp, n_embd}, 0); + layer.ffn_up_shexp = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP, "weight", i), { n_embd, n_ff_shexp}, 0); + } + } break; + case LLM_ARCH_PHI2: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - llama_lora_weight * get_weight(struct ggml_tensor * w) { - std::string name(w->name); - auto pos = ab_map.find(name); - if (ab_map.find(name) != ab_map.end()) { - return &pos->second; - } - return nullptr; - } - - ~llama_lora_adapter() { - auto pos = base_model->lora_adapters.find(this); - if (pos != base_model->lora_adapters.end()) { - base_model->lora_adapters.erase(pos); - } - } -}; - -static int llama_get_device_count(const llama_model & model) { - return (int) model.devices.size(); -} - -template -static bool buft_supported(ggml_backend_buffer_type_t buft, ggml_backend_dev_t dev, F & fn) { - ggml_init_params params = { - /*.mem_size =*/ ggml_tensor_overhead()*8, - /*.mem_buffer =*/ NULL, - /*.no_alloc =*/ true, - }; - ggml_context_ptr ctx { ggml_init(params) }; - if (!ctx) { - throw std::runtime_error(format("failed to create ggml context")); - } - - ggml_backend_buffer_ptr buf { ggml_backend_buft_alloc_buffer(buft, 0) }; - ggml_tensor * op_tensor = fn(ctx.get()); - for (int i = 0; i < GGML_MAX_SRC; i++) { - if (op_tensor->src[i] != nullptr) { - assert(op_tensor->src[i]->buffer == nullptr); - op_tensor->src[i]->buffer = buf.get(); - } - } - bool op_supported = ggml_backend_dev_supports_op(dev, op_tensor); - - return op_supported; -} + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); + model.output_b = create_tensor(tn(LLM_TENSOR_OUTPUT, "bias"), {n_vocab}, 0); -template -static ggml_backend_buffer_type_t select_buft(const llama_model::buft_list_t & buft_list, const F & fn) { - for (const auto & cur : buft_list) { - ggml_backend_dev_t cur_dev = cur.first; - ggml_backend_buffer_type_t cur_buft = cur.second; - if (buft_supported(cur_buft, cur_dev, fn)) { - return cur_buft; - } - } - throw std::runtime_error(format("no suitable buffer type found")); -} + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; -// -// kv cache helpers -// + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); -static bool llama_kv_cache_init( - struct llama_kv_cache & cache, - const llama_context * ctx, - ggml_type type_k, - ggml_type type_v, - uint32_t kv_size, - bool offload) { - const llama_model & model = ctx->model; - const llama_cparams & cparams = ctx->cparams; + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); - const struct llama_hparams & hparams = model.hparams; + if (layer.wqkv == nullptr) { + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, 0); - const int64_t n_layer = hparams.n_layer; + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, 0); - cache.has_shift = false; + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, 0); + } - cache.recurrent = llama_model_is_recurrent(&model); - cache.v_trans = !cache.recurrent && !cparams.flash_attn; + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0); - cache.head = 0; - cache.size = kv_size; - cache.used = 0; + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, 0); - cache.type_k = type_k; - cache.type_v = type_v; + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, 0); + } + } break; + case LLM_ARCH_PHI3: + { + const int64_t n_embd_head = n_embd / n_head; - cache.cells.clear(); - cache.cells.resize(kv_size); + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), { n_embd, n_vocab }, 0); - // create a context for each buffer type - std::map ctx_map; - auto ctx_for_buft = [&](ggml_backend_buffer_type_t buft) -> ggml_context * { - auto it = ctx_map.find(buft); - if (it == ctx_map.end()) { - struct ggml_init_params params = { - /*.mem_size =*/ size_t(2u*n_layer*ggml_tensor_overhead()), - /*.mem_buffer =*/ NULL, - /*.no_alloc =*/ true, - }; - ggml_context * ctx = ggml_init(params); - if (!ctx) { - return nullptr; - } - ctx_map[buft] = ctx; - cache.ctxs.emplace_back(ctx); - return ctx; - } - return it->second; - }; + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), { n_embd }, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), { n_embd, n_vocab }, 0); - cache.k_l.reserve(n_layer); - cache.v_l.reserve(n_layer); + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - for (int i = 0; i < (int) n_layer; i++) { - const uint32_t n_embd_k_gqa = hparams.n_embd_k_gqa(i) + hparams.n_embd_k_s(); - const uint32_t n_embd_v_gqa = hparams.n_embd_v_gqa(i) + hparams.n_embd_v_s(); + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), { n_embd }, 0); - ggml_backend_buffer_type_t buft; - if (offload) { - auto * dev = model.dev_layer.at(i).dev; - buft = ggml_backend_dev_buffer_type(dev); - } else { - buft = ggml_backend_cpu_buffer_type(); - } - ggml_context * ctx = ctx_for_buft(buft); + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), { n_embd, n_embd + 2 * n_embd_gqa }, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), { n_embd, n_embd }, 0); - if (!ctx) { - LLAMA_LOG_ERROR("%s: failed to create ggml context for kv cache\n", __func__); - return false; - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), { n_embd }, 0); - ggml_tensor * k = ggml_new_tensor_1d(ctx, type_k, n_embd_k_gqa*kv_size); - ggml_tensor * v = ggml_new_tensor_1d(ctx, type_v, n_embd_v_gqa*kv_size); - ggml_format_name(k, "cache_k_l%d", i); - ggml_format_name(v, "cache_v_l%d", i); - cache.k_l.push_back(k); - cache.v_l.push_back(v); - } + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd }, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), { n_embd, 2 * n_ff }, 0); - // allocate tensors and initialize the buffers to avoid NaNs in the padding - for (auto it : ctx_map) { - auto * buft = it.first; - auto * ctx = it.second; + layer.rope_long = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG, "weight", i), { n_embd_head/2 }, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0)); + layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), { n_embd_head/2 }, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0)); + } + } break; + case LLM_ARCH_PLAMO: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - ggml_backend_buffer_t buf = ggml_backend_alloc_ctx_tensors_from_buft(ctx, buft); - if (!buf) { - LLAMA_LOG_ERROR("%s: failed to allocate buffer for kv cache\n", __func__); - return false; - } - ggml_backend_buffer_clear(buf, 0); - LLAMA_LOG_INFO("%s: %10s KV buffer size = %8.2f MiB\n", __func__, ggml_backend_buffer_name(buf), ggml_backend_buffer_get_size(buf)/1024.0/1024.0); - cache.bufs.emplace_back(buf); - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - return true; -} + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; -// a structure holds information about the slot found in llama_kv_cache_find_slot -struct llama_kv_cache_slot_info { - std::pair boundaries; // slot boundaries [begin, end) - bool found = false; // the slot was found + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - explicit llama_kv_cache_slot_info(bool found_) : found{found_} {} - llama_kv_cache_slot_info(uint32_t begin, uint32_t end) : boundaries{begin, end}, found{true} {} + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - operator bool() const { return found; } -}; -static const llama_kv_cache_slot_info llama_kv_cache_slot_info_failed{false}; - -// find an empty slot of size "n_tokens" in the cache -// updates the cache head -// returns a structure holding information about the slot found -// Note: On success, it's important that cache.head points -// to the first cell of the slot. -static struct llama_kv_cache_slot_info llama_kv_cache_find_slot( - struct llama_kv_cache & cache, - const struct llama_ubatch & batch) { - const uint32_t n_tokens = batch.n_tokens; - const uint32_t n_seqs = batch.n_seqs; - const uint32_t n_seq_tokens = batch.n_seq_tokens; - - if (cache.recurrent) { - // For recurrent state architectures (like Mamba or RWKV), - // each cache cell can store the state for a whole sequence. - // A slot should be always be contiguous. - - // can only process batches with an equal number of new tokens in each sequence - GGML_ASSERT(batch.equal_seqs); - - int32_t min = cache.size - 1; - int32_t max = 0; - - // everything should fit if all seq_ids are smaller than the max - for (uint32_t s = 0; s < n_seqs; ++s) { - const uint32_t n_seq_id = batch.n_seq_id[s]; - for (uint32_t j = 0; j < n_seq_id; ++j) { - const llama_seq_id seq_id = batch.seq_id[s][j]; - - if (seq_id < 0 || (uint32_t) seq_id >= cache.size) { - // too big seq_id - // TODO: would it be possible to resize the cache instead? - LLAMA_LOG_ERROR("%s: seq_id=%d >= n_seq_max=%d Try using a bigger --parallel value\n", __func__, seq_id, cache.size); - return llama_kv_cache_slot_info_failed; - } - if (j > 0) { - llama_kv_cell & seq = cache.cells[seq_id]; - if (seq.tail >= 0) { - llama_kv_cell & cell = cache.cells[seq.tail]; - // clear cells from seq_ids that become shared - // (should not normally happen, but let's handle it anyway) - cell.seq_id.erase(seq_id); - seq.tail = -1; - if (cell.seq_id.empty()) { - cell.pos = -1; - cell.src = -1; - cache.used -= 1; - } + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); } - } - } - } + } break; + case LLM_ARCH_GPT2: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); + model.pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD, "weight"), {n_embd, n_ctx_train}, 0); -#ifndef NDEBUG - { - std::vector tails_verif; - tails_verif.assign(cache.size, -1); - for (uint32_t i = 0; i < cache.size; ++i) { - llama_kv_cell & cell = cache.cells[i]; - for (llama_seq_id seq_id : cell.seq_id) { - if (tails_verif[seq_id] != -1) { - LLAMA_LOG_ERROR("%s: duplicate tail for seq_id %d in cell %d and %d\n", __func__, seq_id, i, tails_verif[seq_id]); - } - tails_verif[seq_id] = i; - } - } - for (uint32_t i = 0; i < cache.size; ++i) { - if (tails_verif[i] != cache.cells[i].tail) { - LLAMA_LOG_ERROR("%s: wrong tail for seq_id %d, (%d instead of %d)\n", __func__, i, cache.cells[i].tail, tails_verif[i]); - } - } - } -#endif + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - // find next empty cell - uint32_t next_empty_cell = cache.head; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - for (uint32_t i = 0; i < cache.size; ++i) { - if (next_empty_cell >= cache.size) { next_empty_cell -= cache.size; } - llama_kv_cell & cell = cache.cells[next_empty_cell]; - if (cell.is_empty()) { break; } - next_empty_cell += 1; - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); - // find usable cell range - for (uint32_t s = 0; s < n_seqs; ++s) { - const llama_seq_id seq_id = batch.seq_id[s][0]; - llama_kv_cell & seq_meta = cache.cells[seq_id]; - bool has_cell = false; - if (seq_meta.tail >= 0) { - llama_kv_cell & cell = cache.cells[seq_meta.tail]; - GGML_ASSERT(cell.has_seq_id(seq_id)); - // does this seq_id "own" the cell? - if (cell.seq_id.size() == 1) { has_cell = true; } - } - if (!has_cell) { - llama_kv_cell & empty_cell = cache.cells[next_empty_cell]; - GGML_ASSERT(empty_cell.is_empty()); - // copy old tail into the empty cell - if (seq_meta.tail >= 0) { - llama_kv_cell & orig_cell = cache.cells[seq_meta.tail]; - empty_cell.pos = orig_cell.pos; - empty_cell.src = orig_cell.src; - orig_cell.seq_id.erase(seq_id); - empty_cell.seq_id.insert(seq_id); // will be overwritten - } - seq_meta.tail = next_empty_cell; - // find next empty cell - if (s + 1 < n_seqs) { - next_empty_cell += 1; - for (uint32_t i = 0; i < cache.size; ++i) { - if (next_empty_cell >= cache.size) { next_empty_cell -= cache.size; } - llama_kv_cell & cell = cache.cells[next_empty_cell]; - if (cell.is_empty()) { break; } - next_empty_cell += 1; - } - } - } - if (min > seq_meta.tail) { min = seq_meta.tail; } - if (max < seq_meta.tail) { max = seq_meta.tail; } - } + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0); + layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, 0); - // gather and re-order - for (uint32_t s = 0; s < n_seqs; ++s) { - int32_t dst_id = s + min; - int32_t src_id = cache.cells[batch.seq_id[s][0]].tail; - if (dst_id != src_id) { - llama_kv_cell & dst_cell = cache.cells[dst_id]; - llama_kv_cell & src_cell = cache.cells[src_id]; - - std::swap(dst_cell.pos, src_cell.pos); - std::swap(dst_cell.src, src_cell.src); - std::swap(dst_cell.seq_id, src_cell.seq_id); - - // swap tails (assuming they NEVER overlap) - for (const llama_seq_id seq_id : src_cell.seq_id) { - cache.cells[seq_id].tail = src_id; - } - for (const llama_seq_id seq_id : dst_cell.seq_id) { - cache.cells[seq_id].tail = dst_id; - } - } - } + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0); - // update the pos of the used seqs - for (uint32_t s = 0; s < n_seqs; ++s) { - const llama_pos last_pos = batch.pos[n_seq_tokens * s + n_seq_tokens - 1]; - int32_t cell_id = s + min; - llama_kv_cell & cell = cache.cells[cell_id]; - - if (cell.pos >= 0 && last_pos != cell.pos + (llama_pos) n_seq_tokens) { - // What should happen when the pos backtracks or skips a value? - // Clearing the state mid-batch would require special-casing which isn't done. - LLAMA_LOG_WARN("%s: non-consecutive token position %d after %d for sequence %d with %u new tokens\n", - __func__, last_pos, cell.pos, batch.seq_id[s][0], n_seq_tokens); - } - cell.pos = last_pos; - cell.seq_id.clear(); - for (int32_t j = 0; j < batch.n_seq_id[s]; ++j) { - const llama_seq_id seq_id = batch.seq_id[s][j]; - cell.seq_id.insert(seq_id); - cache.cells[seq_id].tail = cell_id; - } - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), {n_embd}, 0); - // allow getting the range of used cells, from head to head + n - cache.head = min; - cache.n = max - min + 1; - cache.used = std::count_if(cache.cells.begin(), cache.cells.end(), - [](const llama_kv_cell& cell){ return !cell.is_empty(); }); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, 0); - // sanity check - return llama_kv_cache_slot_info(cache.n >= n_seqs); - } - // otherwise, one cell per token. + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, 0); + } + } break; + case LLM_ARCH_CODESHELL: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - if (n_tokens > cache.size) { - LLAMA_LOG_ERROR("%s: n_tokens=%d > cache.size=%d\n", __func__, n_tokens, cache.size); - return llama_kv_cache_slot_info_failed; - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - uint32_t n_tested = 0; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - while (true) { - if (cache.head + n_tokens > cache.size) { - n_tested += cache.size - cache.head; - cache.head = 0; - continue; - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); - bool found = true; - for (uint32_t i = 0; i < n_tokens; i++) { - if (cache.cells[cache.head + i].pos >= 0) { - found = false; - cache.head += i + 1; - n_tested += i + 1; - break; - } - } + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0); + layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, 0); - if (found) { - break; - } + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0); - if (n_tested >= cache.size) { - //LLAMA_LOG_ERROR("%s: failed to find a slot for %d tokens\n", __func__, n_tokens); - return llama_kv_cache_slot_info_failed; - } - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), {n_embd}, 0); - for (uint32_t s = 0; s < n_seqs; s++) { - for (uint32_t i = 0; i < n_seq_tokens; ++i) { - uint32_t k = s*n_seq_tokens + i; - cache.cells[cache.head + k].pos = batch.pos[k]; + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, 0); - for (int32_t j = 0; j < batch.n_seq_id[s]; j++) { - cache.cells[cache.head + k].seq_id.insert(batch.seq_id[s][j]); - } - } - } + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, 0); + } + } break; + case LLM_ARCH_ORION: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - cache.used += n_tokens; + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - return llama_kv_cache_slot_info(cache.head, cache.head + n_tokens); -} + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; -// find how many cells are currently in use -static uint32_t llama_kv_cache_cell_max(const struct llama_kv_cache & cache) { - for (uint32_t i = cache.size; i > 0; --i) { - const llama_kv_cell & cell = cache.cells[i - 1]; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); - if (cell.pos >= 0 && !cell.is_empty()) { - return i; - } - } + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - return 0; -} + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), {n_embd}, 0); -static void llama_kv_cache_clear(struct llama_kv_cache & cache) { - for (int32_t i = 0; i < (int32_t) cache.size; ++i) { - cache.cells[i].pos = -1; - cache.cells[i].seq_id.clear(); - cache.cells[i].src = -1; - cache.cells[i].tail = -1; - } - cache.head = 0; - cache.used = 0; + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_INTERNLM2: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - for (auto & buf : cache.bufs) { - ggml_backend_buffer_clear(buf.get(), 0); - } -} + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); -static bool llama_kv_cache_seq_rm( - struct llama_kv_cache & cache, - llama_seq_id seq_id, - llama_pos p0, - llama_pos p1) { - uint32_t new_head = cache.size; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - if (p0 < 0) p0 = 0; - if (p1 < 0) p1 = std::numeric_limits::max(); + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + // layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0); + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); - // models like Mamba or RWKV can't have a state partially erased - if (cache.recurrent) { - if (seq_id >= (int64_t) cache.size) { - // could be fatal - return false; - } - if (0 <= seq_id) { - int32_t & tail_id = cache.cells[seq_id].tail; - if (tail_id >= 0) { - const llama_kv_cell & cell = cache.cells[tail_id]; - // partial intersection is invalid - if ((0 < p0 && p0 <= cell.pos) || (0 < p1 && p1 <= cell.pos)) { - return false; - } - // invalidate tails which will be cleared - if (p0 <= cell.pos && cell.pos < p1) { - tail_id = -1; - } - } - } else { - // seq_id is negative, then the range should include everything or nothing - if (p0 != p1 && (p0 != 0 || p1 != std::numeric_limits::max())) { - return false; - } - } - } + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_GEMMA: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - for (uint32_t i = 0; i < cache.size; ++i) { - if (cache.cells[i].pos >= p0 && cache.cells[i].pos < p1) { - if (seq_id < 0) { - cache.cells[i].seq_id.clear(); - } else if (cache.cells[i].has_seq_id(seq_id)) { - cache.cells[i].seq_id.erase(seq_id); - } else { - continue; - } - if (cache.cells[i].is_empty()) { - // keep count of the number of used cells - if (cache.cells[i].pos >= 0) cache.used--; + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); // same as tok_embd, duplicated to allow offloading - cache.cells[i].pos = -1; - cache.cells[i].src = -1; - if (new_head == cache.size) new_head = i; - } - } - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - // If we freed up a slot, set head to it so searching can start there. - if (new_head != cache.size && new_head < cache.head) cache.head = new_head; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - return true; -} + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_head_k * n_head}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0); -static void llama_kv_cache_seq_cp( - struct llama_kv_cache & cache, - llama_seq_id seq_id_src, - llama_seq_id seq_id_dst, - llama_pos p0, - llama_pos p1) { - if (p0 < 0) p0 = 0; - if (p1 < 0) p1 = std::numeric_limits::max(); - - if (cache.recurrent) { - if ((uint32_t) seq_id_dst < cache.size && (uint32_t) seq_id_src < cache.size) { - llama_kv_cell & tail_src = cache.cells[seq_id_src]; - llama_kv_cell & tail_dst = cache.cells[seq_id_dst]; - if (tail_dst.tail >= 0) { - // clear destination seq_id if it wasn't empty - llama_kv_cell & cell_dst = cache.cells[tail_dst.tail]; - - cell_dst.seq_id.erase(seq_id_dst); - tail_dst.tail = -1; - if (cell_dst.seq_id.empty()) { - cell_dst.pos = -1; - cell_dst.delta = -1; - cell_dst.src = -1; - cache.used -= 1; - } - } - if (tail_src.tail >= 0) { - llama_kv_cell & cell_src = cache.cells[tail_src.tail]; + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + } + } break; + case LLM_ARCH_GEMMA2: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - cell_src.seq_id.insert(seq_id_dst); - tail_dst.tail = tail_src.tail; - } - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); // same as tok_embd, duplicated to allow offloading - return; - } - // otherwise, this is the KV cache of a Transformer-like model + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - cache.head = 0; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - for (uint32_t i = 0; i < cache.size; ++i) { - if (cache.cells[i].has_seq_id(seq_id_src) && cache.cells[i].pos >= p0 && cache.cells[i].pos < p1) { - cache.cells[i].seq_id.insert(seq_id_dst); - } - } -} + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_head_k * n_head}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0); + layer.attn_post_norm = create_tensor(tn(LLM_TENSOR_ATTN_POST_NORM, "weight", i), {n_embd}, 0); -static void llama_kv_cache_seq_keep(struct llama_kv_cache & cache, llama_seq_id seq_id) { - uint32_t new_head = cache.size; + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_post_norm = create_tensor(tn(LLM_TENSOR_FFN_POST_NORM, "weight", i), {n_embd}, 0); + } + } break; + case LLM_ARCH_STARCODER2: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - for (uint32_t i = 0; i < cache.size; ++i) { - if (cache.recurrent && (llama_seq_id) i != seq_id) { - cache.cells[i].tail = -1; - } - if (!cache.cells[i].has_seq_id(seq_id)) { - if (cache.cells[i].pos >= 0) cache.used--; - cache.cells[i].pos = -1; - cache.cells[i].src = -1; - cache.cells[i].seq_id.clear(); - if (new_head == cache.size) new_head = i; - } else { - cache.cells[i].seq_id.clear(); - cache.cells[i].seq_id.insert(seq_id); - } - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); - // If we freed up a slot, set head to it so searching can start there. - if (new_head != cache.size && new_head < cache.head) cache.head = new_head; -} + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); + // if output is NULL, init from the input tok embed + if (model.output == NULL) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); + } -static void llama_kv_cache_seq_add( - struct llama_kv_cache & cache, - llama_seq_id seq_id, - llama_pos p0, - llama_pos p1, - llama_pos delta) { - uint32_t new_head = cache.size; - - if (p0 < 0) p0 = 0; - if (p1 < 0) p1 = std::numeric_limits::max(); - // If there is no range then return early to avoid looping over the cache. - if (p0 == p1) return; - - if (cache.recurrent) { - // for Mamba-like or RWKV models, only the pos needs to be shifted - if (0 <= seq_id && seq_id < (int64_t) cache.size) { - const int32_t tail_id = cache.cells[seq_id].tail; - if (tail_id >= 0) { - llama_kv_cell & cell = cache.cells[tail_id]; - if (cell.has_seq_id(seq_id) && p0 <= cell.pos && cell.pos < p1) { - cell.pos += delta; - } - } - } - return; - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - for (uint32_t i = 0; i < cache.size; ++i) { - if (cache.cells[i].has_seq_id(seq_id) && cache.cells[i].pos >= p0 && cache.cells[i].pos < p1) { - cache.has_shift = true; - cache.cells[i].pos += delta; - cache.cells[i].delta += delta; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); - if (cache.cells[i].pos < 0) { - if (!cache.cells[i].is_empty()) { - cache.used--; - } - cache.cells[i].pos = -1; - cache.cells[i].seq_id.clear(); - if (new_head == cache.size) { - new_head = i; - } - } - } - } + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - // If we freed up a slot, set head to it so searching can start there. - // Otherwise we just start the next search from the beginning. - cache.head = new_head != cache.size ? new_head : 0; -} + // optional bias tensors + layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, 0); + layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, 0); + layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, 0); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0); -static void llama_kv_cache_seq_div( - struct llama_kv_cache & cache, - llama_seq_id seq_id, - llama_pos p0, - llama_pos p1, - int d) { - if (p0 < 0) p0 = 0; - if (p1 < 0) p1 = std::numeric_limits::max(); - // If there is no range then return early to avoid looping over the cache. - if (p0 == p1) return; - - if (cache.recurrent) { - // for Mamba-like or RWKV models, only the pos needs to be changed - if (0 <= seq_id && seq_id < (int64_t) cache.size) { - const int32_t tail_id = cache.cells[seq_id].tail; - if (tail_id >= 0) { - llama_kv_cell & cell = cache.cells[tail_id]; - if (cell.has_seq_id(seq_id) && p0 <= cell.pos && cell.pos < p1) { - cell.pos /= d; - } - } - } - return; - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), {n_embd}, 0); - for (uint32_t i = 0; i < cache.size; ++i) { - if (cache.cells[i].has_seq_id(seq_id) && cache.cells[i].pos >= p0 && cache.cells[i].pos < p1) { - cache.has_shift = true; + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); - { - llama_pos p_old = cache.cells[i].pos; - cache.cells[i].pos /= d; - cache.cells[i].delta += cache.cells[i].pos - p_old; - } - } - } -} + // optional bias tensors + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, 0); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP , "bias", i), { n_ff}, 0); + } + } break; + case LLM_ARCH_MAMBA: + { + const int64_t d_conv = hparams.ssm_d_conv; + const int64_t d_inner = hparams.ssm_d_inner; + const int64_t d_state = hparams.ssm_d_state; + const int64_t dt_rank = hparams.ssm_dt_rank; -static llama_pos llama_kv_cache_seq_pos_max(struct llama_kv_cache & cache, llama_seq_id seq_id) { - llama_pos result = 0; + // only an expansion factor of 2 is supported for now + if (2 * n_embd != d_inner) { + throw std::runtime_error("only an expansion factor of 2 is supported for now"); + } - for (uint32_t i = 0; i < cache.size; ++i) { - if (cache.cells[i].has_seq_id(seq_id)) { - result = std::max(result, cache.cells[i].pos); - } - } + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - return result; -} + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); -static void llama_kv_cache_defrag(struct llama_kv_cache & cache) { - if (!cache.recurrent) { - cache.do_defrag = true; - } -} + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); + // if output is NULL, init from the input tok embed, duplicated to allow offloading + if (model.output == NULL) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); + } -static uint32_t llama_kv_cache_get_padding(const struct llama_cparams & cparams) { - // the FA kernels require padding to avoid extra runtime boundary checks - return cparams.flash_attn ? 256u : 32u; -} + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; -// saves the kv_cache state for future recovery. -// used to rollback llama_kv_cache_find_slot changes. -struct llama_kv_slot_restorer { - struct llama_kv_cache_state { - uint32_t head = 0; - uint32_t n = 0; - } old_state; + // norm + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - // for non-recurrent models only - // list of slots to restore - std::vector> slot_boundaries; + layer.ssm_in = create_tensor(tn(LLM_TENSOR_SSM_IN, "weight", i), {n_embd, 2*d_inner}, 0); - bool do_restore = false; + layer.ssm_conv1d = create_tensor(tn(LLM_TENSOR_SSM_CONV1D, "weight", i), {d_conv, d_inner}, 0); + layer.ssm_conv1d_b = create_tensor(tn(LLM_TENSOR_SSM_CONV1D, "bias", i), {d_inner}, 0); - explicit llama_kv_slot_restorer(const struct llama_kv_cache & cache) { - old_state.head = cache.head; - old_state.n = cache.n; - } + layer.ssm_x = create_tensor(tn(LLM_TENSOR_SSM_X, "weight", i), {d_inner, dt_rank + 2*d_state}, 0); - // saves a slot information for future restoration - void save(const struct llama_kv_cache_slot_info & slot) { - if (slot) { - do_restore = true; - if (slot.boundaries.first != slot.boundaries.second) { - slot_boundaries.push_back(slot.boundaries); - } - } - } + layer.ssm_dt = create_tensor(tn(LLM_TENSOR_SSM_DT, "weight", i), {dt_rank, d_inner}, 0); + layer.ssm_dt_b = create_tensor(tn(LLM_TENSOR_SSM_DT, "bias", i), {d_inner}, 0); - // must be explicitly called to restore the kv_cache state - // and rollback changes from all llama_kv_cache_find_slot calls - void restore(struct llama_kv_cache & cache) { - if (do_restore) { - cache.head = old_state.head; - cache.n = old_state.n; + // no "weight" suffix for these + layer.ssm_a = create_tensor(tn(LLM_TENSOR_SSM_A, i), {d_state, d_inner}, 0); + layer.ssm_d = create_tensor(tn(LLM_TENSOR_SSM_D, i), {d_inner}, 0); - if (cache.recurrent) { // recurrent models like Mamba or RWKV can't have a state partially erased - llama_kv_cache_seq_rm(cache, -1, -1, -1); - } else { - for (auto & slot : slot_boundaries) { - llama_kv_cache_seq_rm(cache, -1, slot.first, slot.second); - } - } - } - } -}; + // out_proj + layer.ssm_out = create_tensor(tn(LLM_TENSOR_SSM_OUT, "weight", i), {d_inner, n_embd}, 0); + } + } break; + case LLM_ARCH_XVERSE: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); -// -// model loading and saving -// + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); -enum llama_fver { - GGUF_FILE_VERSION_V1 = 1, - GGUF_FILE_VERSION_V2 = 2, - GGUF_FILE_VERSION_V3 = 3, -}; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; -static const char * llama_file_version_name(llama_fver version) { - switch (version) { - case GGUF_FILE_VERSION_V1: return "GGUF V1 (support until nov 2023)"; - case GGUF_FILE_VERSION_V2: return "GGUF V2"; - case GGUF_FILE_VERSION_V3: return "GGUF V3 (latest)"; - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - return "unknown"; -} + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); -static std::string llama_format_tensor_shape(const std::vector & ne) { - char buf[256]; - snprintf(buf, sizeof(buf), "%5" PRId64, ne.at(0)); - for (size_t i = 1; i < ne.size(); i++) { - snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), ", %5" PRId64, ne.at(i)); - } - return buf; -} + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_COMMAND_R: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); -static std::string llama_format_tensor_shape(const struct ggml_tensor * t) { - char buf[256]; - snprintf(buf, sizeof(buf), "%5" PRId64, t->ne[0]); - for (int i = 1; i < GGML_MAX_DIMS; i++) { - snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), ", %5" PRId64, t->ne[i]); - } - return buf; -} + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + // init output from the input tok embed + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); -namespace GGUFMeta { - template - struct GKV_Base_Type { - static constexpr gguf_type gt = gt_; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - static T getter(const gguf_context * ctx, const int kid) { - return gfun(ctx, kid); - } - }; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - template struct GKV_Base; - - template<> struct GKV_Base: GKV_Base_Type {}; - template<> struct GKV_Base: GKV_Base_Type {}; - template<> struct GKV_Base: GKV_Base_Type {}; - template<> struct GKV_Base: GKV_Base_Type {}; - template<> struct GKV_Base: GKV_Base_Type {}; - template<> struct GKV_Base: GKV_Base_Type {}; - template<> struct GKV_Base: GKV_Base_Type {}; - template<> struct GKV_Base: GKV_Base_Type {}; - template<> struct GKV_Base: GKV_Base_Type {}; - template<> struct GKV_Base: GKV_Base_Type {}; - template<> struct GKV_Base: GKV_Base_Type {}; - template<> struct GKV_Base: GKV_Base_Type {}; - - template<> struct GKV_Base { - static constexpr gguf_type gt = GGUF_TYPE_STRING; - - static std::string getter(const gguf_context * ctx, const int kid) { - return gguf_get_val_str(ctx, kid); - } - }; + if (n_layer >= 64){ + layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k, n_head}, 0); + layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k, n_head_kv}, 0); + } - struct ArrayInfo { - const gguf_type gt; - const size_t length; - const void * data; - }; + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - template<> struct GKV_Base { - public: - static constexpr gguf_type gt = GGUF_TYPE_ARRAY; - static ArrayInfo getter(const gguf_context *ctx, const int k) { - return ArrayInfo { - gguf_get_arr_type(ctx, k), - size_t(gguf_get_arr_n(ctx, k)), - gguf_get_arr_data(ctx, k), - }; - } - }; + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_COHERE2: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), { n_embd, n_vocab }, 0); - template - class GKV : public GKV_Base { - GKV() = delete; + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), { n_embd }, 0); + // init output from the input tok embed + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), { n_embd, n_vocab }, + llama_model_loader::TENSOR_DUPLICATED); - public: - static T get_kv(const gguf_context * ctx, const int k) { - const enum gguf_type kt = gguf_get_kv_type(ctx, k); + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - if (kt != GKV::gt) { - throw std::runtime_error(format("key %s has wrong type %s but expected type %s", - gguf_get_key(ctx, k), gguf_type_name(kt), gguf_type_name(GKV::gt))); - } - return GKV::getter(ctx, k); - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), { n_embd }, 0); - static const char * override_type_to_str(const llama_model_kv_override_type ty) { - switch (ty) { - case LLAMA_KV_OVERRIDE_TYPE_BOOL: return "bool"; - case LLAMA_KV_OVERRIDE_TYPE_INT: return "int"; - case LLAMA_KV_OVERRIDE_TYPE_FLOAT: return "float"; - case LLAMA_KV_OVERRIDE_TYPE_STR: return "str"; - } - return "unknown"; - } + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), { n_embd, n_embd }, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), { n_embd, n_embd_gqa }, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), { n_embd, n_embd_gqa }, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), { n_embd, n_embd }, 0); - static bool validate_override(const llama_model_kv_override_type expected_type, const struct llama_model_kv_override * ovrd) { - if (!ovrd) { return false; } - if (ovrd->tag == expected_type) { - LLAMA_LOG_INFO("%s: Using metadata override (%5s) '%s' = ", - __func__, override_type_to_str(ovrd->tag), ovrd->key); - switch (ovrd->tag) { - case LLAMA_KV_OVERRIDE_TYPE_BOOL: { - LLAMA_LOG_INFO("%s\n", ovrd->val_bool ? "true" : "false"); - } break; - case LLAMA_KV_OVERRIDE_TYPE_INT: { - LLAMA_LOG_INFO("%" PRId64 "\n", ovrd->val_i64); - } break; - case LLAMA_KV_OVERRIDE_TYPE_FLOAT: { - LLAMA_LOG_INFO("%.6f\n", ovrd->val_f64); - } break; - case LLAMA_KV_OVERRIDE_TYPE_STR: { - LLAMA_LOG_INFO("%s\n", ovrd->val_str); - } break; - default: - // Shouldn't be possible to end up here, but just in case... - throw std::runtime_error( - format("Unsupported attempt to override %s type for metadata key %s\n", - override_type_to_str(ovrd->tag), ovrd->key)); + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), { n_embd, n_ff }, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd }, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), { n_embd, n_ff }, 0); + } } - return true; - } - LLAMA_LOG_WARN("%s: Warning: Bad metadata override type for key '%s', expected %s but got %s\n", - __func__, ovrd->key, override_type_to_str(expected_type), override_type_to_str(ovrd->tag)); - return false; - } - - template - static typename std::enable_if::value, bool>::type - try_override(OT & target, const struct llama_model_kv_override * ovrd) { - if (validate_override(LLAMA_KV_OVERRIDE_TYPE_BOOL, ovrd)) { - target = ovrd->val_bool; - return true; - } - return false; - } - - template - static typename std::enable_if::value && std::is_integral::value, bool>::type - try_override(OT & target, const struct llama_model_kv_override * ovrd) { - if (validate_override(LLAMA_KV_OVERRIDE_TYPE_INT, ovrd)) { - target = ovrd->val_i64; - return true; - } - return false; - } - - template - static typename std::enable_if::value, bool>::type - try_override(T & target, const struct llama_model_kv_override * ovrd) { - if (validate_override(LLAMA_KV_OVERRIDE_TYPE_FLOAT, ovrd)) { - target = ovrd->val_f64; - return true; - } - return false; - } + break; + case LLM_ARCH_OLMO: // adapted from LLM_ARCH_LLAMA with norm params removed + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - template - static typename std::enable_if::value, bool>::type - try_override(T & target, const struct llama_model_kv_override * ovrd) { - if (validate_override(LLAMA_KV_OVERRIDE_TYPE_STR, ovrd)) { - target = ovrd->val_str; - return true; - } - return false; - } + // output + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); + // if output is NULL, init from the input tok embed + if (model.output == NULL) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); + } - static bool set(const gguf_context * ctx, const int k, T & target, const struct llama_model_kv_override * ovrd = nullptr) { - if (try_override(target, ovrd)) { - return true; - } - if (k < 0) { return false; } - target = get_kv(ctx, k); - return true; - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - static bool set(const gguf_context * ctx, const char * key, T & target, const struct llama_model_kv_override * ovrd = nullptr) { - return set(ctx, gguf_find_key(ctx, key), target, ovrd); - } + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - static bool set(const gguf_context * ctx, const std::string & key, T & target, const struct llama_model_kv_override * ovrd = nullptr) { - return set(ctx, key.c_str(), target, ovrd); - } - }; -} + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_OLMO2: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); -using llama_buf_map = std::unordered_map; + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); -static size_t llama_model_max_nodes(const llama_model & model) { - return std::max(8192, model.tensors_by_name.size()*5); -} + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; -struct llama_model_loader { - int n_kv = 0; - int n_tensors = 0; - int n_created = 0; + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd}, 0); + layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd}, 0); + layer.attn_post_norm = create_tensor(tn(LLM_TENSOR_ATTN_POST_NORM, "weight", i), {n_embd}, 0); - uint64_t n_elements = 0; - size_t n_bytes = 0; + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_post_norm = create_tensor(tn(LLM_TENSOR_FFN_POST_NORM, "weight", i), {n_embd}, 0); + } + } break; + case LLM_ARCH_OLMOE: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - bool use_mmap = false; - bool check_tensors; + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - llama_files files; - llama_ftype ftype; - llama_fver fver; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - llama_mmaps mappings; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - // Holds information on a model weight - struct llama_tensor_weight { - uint16_t idx; // source file index - size_t offs; // tensor data offset in the original file + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd}, 0); + layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd}, 0); - ggml_tensor * tensor; + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - llama_tensor_weight(const llama_file * file, uint16_t idx, const struct gguf_context * gguf_ctx, ggml_tensor * tensor) : idx(idx), tensor(tensor) { - const int tensor_idx = gguf_find_tensor(gguf_ctx, ggml_get_name(tensor)); - if (tensor_idx < 0) { - throw std::runtime_error(format("tensor '%s' not found in the model", ggml_get_name(tensor))); - } + layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0); - offs = gguf_get_data_offset(gguf_ctx) + gguf_get_tensor_offset(gguf_ctx, tensor_idx); - if (offs + ggml_nbytes(tensor) < offs || offs + ggml_nbytes(tensor) > file->size) { - throw std::runtime_error(format("tensor '%s' data is not within the file bounds, model is corrupted or incomplete", ggml_get_name(tensor))); - } - } - }; + if (n_expert == 0) { + throw std::runtime_error("n_expert must be > 0"); + } + if (n_expert_used == 0) { + throw std::runtime_error("n_expert_used must be > 0"); + } - // custom comparator to sort weights more nicely by layer - struct weight_name_comparer { - bool operator()(const std::string & a, const std::string & b) const { - int a_layer = -1; - int b_layer = -1; - sscanf(a.c_str(), "blk.%d.", &a_layer); - sscanf(b.c_str(), "blk.%d.", &b_layer); - if (a_layer != b_layer) { - return a_layer < b_layer; - } - return a < b; - } - }; + // MoE branch + layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff, n_expert}, 0); + layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff, n_embd, n_expert}, 0); + layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), {n_embd, n_ff, n_expert}, 0); + } + } break; + case LLM_ARCH_OPENELM: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - std::map weights_map; - std::unordered_map kv_overrides; + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + // init output from the input tok embed + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); - gguf_context_ptr meta; - std::vector contexts; + for (int i = 0; i < n_layer; ++i) { + const int64_t n_head = hparams.n_head(i); + const int64_t n_head_qkv = 2*hparams.n_head_kv(i) + n_head; + const int64_t n_ff = hparams.n_ff(i); - std::string arch_name; - LLM_KV llm_kv = LLM_KV(LLM_ARCH_UNKNOWN); + auto & layer = model.layers[i]; - llama_model_loader(const std::string & fname, bool use_mmap, bool check_tensors, const struct llama_model_kv_override * param_overrides_p) { - int trace = 0; - if (getenv("LLAMA_TRACE")) { - trace = atoi(getenv("LLAMA_TRACE")); - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - if (param_overrides_p != nullptr) { - for (const struct llama_model_kv_override * p = param_overrides_p; p->key[0] != 0; p++) { - kv_overrides.insert({std::string(p->key), *p}); - } - } + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_head_qkv*n_embd_head_k}, 0); + layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k}, 0); + layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_head*n_embd_head_k, n_embd}, 0); - struct ggml_context * ctx = NULL; - struct gguf_init_params params = { - /*.no_alloc = */ true, - /*.ctx = */ &ctx, - }; + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_GPTNEOX: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - meta.reset(gguf_init_from_file(fname.c_str(), params)); - if (!meta) { - throw std::runtime_error(format("%s: failed to load model from %s\n", __func__, fname.c_str())); - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - get_key(llm_kv(LLM_KV_GENERAL_ARCHITECTURE), arch_name, false); - llm_kv = LLM_KV(llm_arch_from_string(arch_name)); + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - files.emplace_back(new llama_file(fname.c_str(), "rb")); - contexts.emplace_back(ctx); + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); - // Save tensors data offset of the main file. - // For subsidiary files, `meta` tensor data offset must not be used, - // so we build a unified tensors index for weights. - for (ggml_tensor * cur = ggml_get_first_tensor(ctx); cur; cur = ggml_get_next_tensor(ctx, cur)) { - std::string tensor_name = std::string(cur->name); - // make sure there is no duplicated tensor names - if (weights_map.find(tensor_name) != weights_map.end()) { - throw std::runtime_error(format("invalid model: tensor '%s' is duplicated", ggml_get_name(cur))); - } - n_elements += ggml_nelements(cur); - n_bytes += ggml_nbytes(cur); - weights_map.emplace(tensor_name, llama_tensor_weight(files.back().get(), 0, meta.get(), cur)); - } - uint16_t n_split = 0; - get_key(llm_kv(LLM_KV_SPLIT_COUNT), n_split, false); - - // Load additional GGML contexts - if (n_split > 1) { - uint16_t idx = 0; - get_key(llm_kv(LLM_KV_SPLIT_NO), idx); - if (idx != 0) { - throw std::runtime_error(format("illegal split file: %d, model must be loaded with the first split", idx)); - } + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0); + layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, 0); - char split_prefix[PATH_MAX] = {0}; - if (!llama_split_prefix(split_prefix, sizeof(split_prefix), fname.c_str(), idx, n_split)) { - throw std::runtime_error(format("invalid split file: %s", fname.c_str())); - } + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0); - if (trace > 0) { - LLAMA_LOG_INFO("%s: loading additional %d GGUFs\n", __func__, n_split); - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), {n_embd}, 0); - char split_path[PATH_MAX] = {0}; - for (idx = 1; idx < n_split; idx++) { - llama_split_path(split_path, sizeof(split_path), split_prefix, idx, n_split); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, 0); - struct gguf_init_params split_params = { - /*.no_alloc = */ true, - /*.ctx = */ &ctx, - }; - gguf_context_ptr ctx_gguf { gguf_init_from_file(split_path, split_params) }; - if (!ctx_gguf) { - throw std::runtime_error(format("%s: failed to load GGUF split from %s\n", __func__, split_path)); - } + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, 0); + } + } break; + case LLM_ARCH_ARCTIC: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - files.emplace_back(new llama_file(split_path, "rb")); - contexts.emplace_back(ctx); + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); - // Save tensors data offset info of the shard. - for (ggml_tensor * cur = ggml_get_first_tensor(ctx); cur; cur = ggml_get_next_tensor(ctx, cur)) { - std::string tensor_name = std::string(cur->name); - // make sure there is no duplicated tensor names - if (weights_map.find(tensor_name) != weights_map.end()) { - throw std::runtime_error(format("invalid model: tensor '%s' is duplicated", ggml_get_name(cur))); + // if output is NULL, init from the input tok embed + if (model.output == NULL) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); } - n_elements += ggml_nelements(cur); - n_bytes += ggml_nbytes(cur); - weights_map.emplace(tensor_name, llama_tensor_weight(files.back().get(), idx, ctx_gguf.get(), cur)); - } - } - get_key(llm_kv(LLM_KV_SPLIT_TENSORS_COUNT), n_tensors); - - // sanity check - { - const int n_tensors_loaded = (int) weights_map.size(); - if (n_tensors != n_tensors_loaded) { - throw std::runtime_error(format("corrupted model: %d tensors expected but %d found", n_tensors, n_tensors_loaded)); - } - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - LLAMA_LOG_INFO("%s: additional %d GGUFs metadata loaded.\n", __func__, n_split - 1); - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - n_kv = gguf_get_n_kv(meta.get()); - n_tensors = weights_map.size(); + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - fver = (enum llama_fver) gguf_get_version(meta.get()); + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - LLAMA_LOG_INFO("%s: loaded meta data with %d key-value pairs and %d tensors from %s (version %s)\n", - __func__, n_kv, n_tensors, fname.c_str(), llama_file_version_name(fver)); + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_embd}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_embd, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_embd}, 0); - // determine file type based on the number of tensors for each quantization and print meta data - // TODO: make optional - { - std::map n_type; + layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0); + layer.ffn_norm_exps = create_tensor(tn(LLM_TENSOR_FFN_NORM_EXPS, "weight", i), {n_embd}, 0); + layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff, n_expert}, false); + layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), { n_ff, n_embd, n_expert}, 0); + layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), {n_embd, n_ff, n_expert}, 0); + } + } break; + case LLM_ARCH_DEEPSEEK: + { - uint32_t n_type_max = 0; - enum ggml_type type_max = GGML_TYPE_F32; + const int64_t n_ff_exp = hparams.n_ff_exp; + const int64_t n_expert_shared = hparams.n_expert_shared; - for (const auto & it : weights_map) { - const llama_tensor_weight & w = it.second; - const ggml_tensor * tensor = w.tensor; + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - enum ggml_type type = tensor->type; + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - n_type[type]++; + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - if (n_type_max < n_type[type]) { - n_type_max = n_type[type]; - type_max = type; - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - if (trace > 0) { - const uint16_t sid = w.idx; - LLAMA_LOG_INFO("%s: - tensor split %2d: %32s %-8s [ %s ]\n", __func__, sid, ggml_get_name(tensor), ggml_type_name(type), llama_format_tensor_shape(tensor).c_str()); - } - } + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - switch (type_max) { - case GGML_TYPE_F32: ftype = LLAMA_FTYPE_ALL_F32; break; - case GGML_TYPE_F16: ftype = LLAMA_FTYPE_MOSTLY_F16; break; - case GGML_TYPE_BF16: ftype = LLAMA_FTYPE_MOSTLY_BF16; break; - case GGML_TYPE_Q4_0: ftype = LLAMA_FTYPE_MOSTLY_Q4_0; break; - case GGML_TYPE_Q4_1: ftype = LLAMA_FTYPE_MOSTLY_Q4_1; break; - case GGML_TYPE_Q5_0: ftype = LLAMA_FTYPE_MOSTLY_Q5_0; break; - case GGML_TYPE_Q5_1: ftype = LLAMA_FTYPE_MOSTLY_Q5_1; break; - case GGML_TYPE_Q8_0: ftype = LLAMA_FTYPE_MOSTLY_Q8_0; break; - case GGML_TYPE_Q2_K: ftype = LLAMA_FTYPE_MOSTLY_Q2_K; break; - case GGML_TYPE_Q3_K: ftype = LLAMA_FTYPE_MOSTLY_Q3_K_M; break; - case GGML_TYPE_Q4_K: ftype = LLAMA_FTYPE_MOSTLY_Q4_K_M; break; - case GGML_TYPE_Q5_K: ftype = LLAMA_FTYPE_MOSTLY_Q5_K_M; break; - case GGML_TYPE_Q6_K: ftype = LLAMA_FTYPE_MOSTLY_Q6_K; break; - case GGML_TYPE_TQ1_0: ftype = LLAMA_FTYPE_MOSTLY_TQ1_0; break; - case GGML_TYPE_TQ2_0: ftype = LLAMA_FTYPE_MOSTLY_TQ2_0; break; - case GGML_TYPE_IQ2_XXS: ftype = LLAMA_FTYPE_MOSTLY_IQ2_XXS; break; - case GGML_TYPE_IQ2_XS: ftype = LLAMA_FTYPE_MOSTLY_IQ2_XS; break; - case GGML_TYPE_IQ2_S: ftype = LLAMA_FTYPE_MOSTLY_IQ2_S; break; - case GGML_TYPE_IQ3_XXS: ftype = LLAMA_FTYPE_MOSTLY_IQ3_XXS; break; - case GGML_TYPE_IQ1_S: ftype = LLAMA_FTYPE_MOSTLY_IQ1_S; break; - case GGML_TYPE_IQ1_M: ftype = LLAMA_FTYPE_MOSTLY_IQ1_M; break; - case GGML_TYPE_IQ4_NL: ftype = LLAMA_FTYPE_MOSTLY_IQ4_NL; break; - case GGML_TYPE_IQ4_XS: ftype = LLAMA_FTYPE_MOSTLY_IQ4_XS; break; - case GGML_TYPE_IQ3_S: ftype = LLAMA_FTYPE_MOSTLY_IQ3_S; break; - default: - { - LLAMA_LOG_WARN("%s: unknown type %s\n", __func__, ggml_type_name(type_max)); - ftype = LLAMA_FTYPE_ALL_F32; - } break; - } + if (i < (int) hparams.n_layer_dense_lead) { + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } else { + layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0); - // this is a way to mark that we have "guessed" the file type - ftype = (llama_ftype) (ftype | LLAMA_FTYPE_GUESSED); + if (n_expert == 0) { + throw std::runtime_error("n_expert must be > 0"); + } + if (n_expert_used == 0) { + throw std::runtime_error("n_expert_used must be > 0"); + } - { - const int kid = gguf_find_key(meta.get(), "general.file_type"); // TODO: use LLM_KV - if (kid >= 0) { - ftype = (llama_ftype) gguf_get_val_u32(meta.get(), kid); - } - } + // MoE branch + layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), { n_embd, n_ff_exp, n_expert}, 0); + layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp, n_embd, n_expert}, 0); + layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), { n_embd, n_ff_exp, n_expert}, 0); - LLAMA_LOG_INFO("%s: Dumping metadata keys/values. Note: KV overrides do not apply in this output.\n", __func__); + // Shared expert branch + layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0); + layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), { n_ff_exp * n_expert_shared, n_embd}, 0); + layer.ffn_up_shexp = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP, "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0); + } + } + } break; + case LLM_ARCH_DEEPSEEK2: + { + const bool is_lite = (hparams.n_layer == 27); - for (int i = 0; i < n_kv; i++) { - const char * name = gguf_get_key(meta.get(), i); - const enum gguf_type type = gguf_get_kv_type(meta.get(), i); - const std::string type_name = - type == GGUF_TYPE_ARRAY - ? format("%s[%s,%d]", gguf_type_name(type), gguf_type_name(gguf_get_arr_type(meta.get(), i)), gguf_get_arr_n(meta.get(), i)) - : gguf_type_name(type); + const int64_t n_embd_head_qk_rope = hparams.n_rot; + const int64_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot; - std::string value = gguf_kv_to_str(meta.get(), i); - const size_t MAX_VALUE_LEN = 40; - if (value.size() > MAX_VALUE_LEN) { - value = format("%s...", value.substr(0, MAX_VALUE_LEN - 3).c_str()); - } - replace_all(value, "\n", "\\n"); - - LLAMA_LOG_INFO("%s: - kv %3d: %42s %-16s = %s\n", __func__, i, name, type_name.c_str(), value.c_str()); - } + const int64_t q_lora_rank = hparams.n_lora_q; + const int64_t kv_lora_rank = hparams.n_lora_kv; - // print type counts - for (auto & kv : n_type) { - if (kv.second == 0) { - continue; - } + const int64_t n_ff_exp = hparams.n_ff_exp; + const int64_t n_expert_shared = hparams.n_expert_shared; - LLAMA_LOG_INFO("%s: - type %4s: %4d tensors\n", __func__, ggml_type_name(kv.first), kv.second); - } - } + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - if (!llama_mmap::SUPPORTED) { - LLAMA_LOG_WARN("%s: mmap is not supported on this platform\n", __func__); - use_mmap = false; - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - this->use_mmap = use_mmap; - this->check_tensors = check_tensors; - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - template - typename std::enable_if::value, bool>::type - get_arr_n(const std::string & key, T & result, const bool required = true) { - const int kid = gguf_find_key(meta.get(), key.c_str()); + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + if (!is_lite) { + layer.attn_q_a_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_A_NORM, "weight", i), {q_lora_rank}, 0); + } - if (kid < 0) { - if (required) { - throw std::runtime_error(format("key not found in model: %s", key.c_str())); - } - return false; - } + layer.attn_kv_a_norm = create_tensor(tn(LLM_TENSOR_ATTN_KV_A_NORM, "weight", i), {kv_lora_rank}, 0); - struct GGUFMeta::ArrayInfo arr_info = - GGUFMeta::GKV::get_kv(meta.get(), kid); + if (!is_lite) { + layer.wq_a = create_tensor(tn(LLM_TENSOR_ATTN_Q_A, "weight", i), {n_embd, q_lora_rank}, 0); + layer.wq_b = create_tensor(tn(LLM_TENSOR_ATTN_Q_B, "weight", i), {q_lora_rank, n_head * n_embd_head_k}, 0); + } else { + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_k_gqa}, 0); + } + layer.wkv_a_mqa = create_tensor(tn(LLM_TENSOR_ATTN_KV_A_MQA, "weight", i), {n_embd, kv_lora_rank + (n_embd_head_qk_rope)}, 0); + layer.wkv_b = create_tensor(tn(LLM_TENSOR_ATTN_KV_B, "weight", i), {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), { n_head * ( n_embd_head_v), n_embd}, 0); - result = arr_info.length; - return true; - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - template - typename std::enable_if::value, bool>::type - get_arr_n(const enum llm_kv kid, T & result, const bool required = true) { - return get_arr_n(llm_kv(kid), result, required); - } + if (i < (int) hparams.n_layer_dense_lead) { + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } else { + layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0); + layer.ffn_exp_probs_b = create_tensor(tn(LLM_TENSOR_FFN_EXP_PROBS_B, "bias", i), {n_expert}, llama_model_loader::TENSOR_NOT_REQUIRED); - template - bool get_arr(const std::string & key, std::vector & result, const bool required = true) { - const int kid = gguf_find_key(meta.get(), key.c_str()); + if (n_expert == 0) { + throw std::runtime_error("n_expert must be > 0"); + } + if (n_expert_used == 0) { + throw std::runtime_error("n_expert_used must be > 0"); + } - if (kid < 0 || gguf_get_kv_type(meta.get(), kid) != GGUF_TYPE_ARRAY) { - if (required) { - throw std::runtime_error(format("array key not found in model: %s", key.c_str())); - } - return false; - } + // MoE branch + layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), { n_embd, n_ff_exp, n_expert}, 0); + layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp, n_embd, n_expert}, 0); + layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), { n_embd, n_ff_exp, n_expert}, 0); - struct GGUFMeta::ArrayInfo arr_info = - GGUFMeta::GKV::get_kv(meta.get(), kid); + // Shared expert branch + layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0); + layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), { n_ff_exp * n_expert_shared, n_embd}, 0); + layer.ffn_up_shexp = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP, "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0); + } + } + } break; + case LLM_ARCH_BITNET: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - switch (arr_info.gt) { - case GGUF_TYPE_FLOAT32: GGML_ASSERT((std::is_same::value)); break; - case GGUF_TYPE_INT32: GGML_ASSERT( - (std::is_same::value) || - (std::is_same::value)); break; - default: - throw std::runtime_error(format("%s is not a float32, int32 array", key.c_str())); - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); - result.resize(arr_info.length); - result.assign((const T*)arr_info.data, (const T *)arr_info.data + arr_info.length); + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - return true; - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_sub_norm = create_tensor(tn(LLM_TENSOR_ATTN_SUB_NORM, "weight", i), {n_embd}, 0); - template - bool get_arr(const std::string & key, std::array & result, const bool required = true) { - const int kid = gguf_find_key(meta.get(), key.c_str()); + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wq_scale = create_tensor(tn(LLM_TENSOR_ATTN_Q, "scale", i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wk_scale = create_tensor(tn(LLM_TENSOR_ATTN_K, "scale", i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv_scale = create_tensor(tn(LLM_TENSOR_ATTN_V, "scale", i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.wo_scale = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "scale", i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED); - if (kid < 0 || gguf_get_kv_type(meta.get(), kid) != GGUF_TYPE_ARRAY) { - if (required) { - throw std::runtime_error(format("array key not found in model: %s", key.c_str())); - } - return false; - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_sub_norm = create_tensor(tn(LLM_TENSOR_FFN_SUB_NORM, "weight", i), {n_ff}, 0); - struct GGUFMeta::ArrayInfo arr_info = - GGUFMeta::GKV::get_kv(meta.get(), kid); + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_gate_scale = create_tensor(tn(LLM_TENSOR_FFN_GATE, "scale", i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); + layer.ffn_down_scale = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "scale", i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_up_scale = create_tensor(tn(LLM_TENSOR_FFN_UP, "scale", i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED); + } + } break; + case LLM_ARCH_T5: + { + const auto n_rel_attn_bkts = hparams.n_rel_attn_bkts; - switch (arr_info.gt) { - case GGUF_TYPE_FLOAT32: GGML_ASSERT((std::is_same::value)); break; - case GGUF_TYPE_INT32: GGML_ASSERT( - (std::is_same::value) || - (std::is_same::value)); break; - default: - throw std::runtime_error(format("%s is not a float32, int32 array", key.c_str())); - } + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - if (arr_info.length > N_MAX) { - throw std::runtime_error(format("array length %u for key %s exceeds max %u", (uint32_t) arr_info.length, key.c_str(), (uint32_t) N_MAX)); - } + // output + model.output_norm_enc = create_tensor(tn(LLM_TENSOR_ENC_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm = create_tensor(tn(LLM_TENSOR_DEC_OUTPUT_NORM, "weight"), {n_embd}, 0); - std::copy((const T*)arr_info.data, (const T *)arr_info.data + arr_info.length, result.begin()); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); + // if output is NULL, init from the input tok embed + if (model.output == NULL) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); + } - return true; - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - template - bool get_arr(const enum llm_kv kid, T & result, const bool required = true) { - return get_arr(llm_kv(kid), result, required); - } + layer.attn_norm_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_rel_b_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_REL_B, "weight", i), {n_head, n_rel_attn_bkts}, llama_model_loader::TENSOR_NOT_REQUIRED); - template - bool get_key(const std::string & key, T & result, const bool required = true) { - auto it = kv_overrides.find(key); + layer.wq_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_Q, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wk_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wv_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0); + layer.wo_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_OUT, "weight", i), {n_embd_v_gqa, n_embd}, 0); - const struct llama_model_kv_override * override = - it != kv_overrides.end() ? &it->second : nullptr; + layer.ffn_norm_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_gate_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_GATE, "weight", i), {n_embd, n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_down_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_UP, "weight", i), {n_embd, n_ff}, 0); - const bool found = GGUFMeta::GKV::set(meta.get(), key, result, override); + layer.attn_norm = create_tensor(tn(LLM_TENSOR_DEC_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_rel_b = create_tensor(tn(LLM_TENSOR_DEC_ATTN_REL_B, "weight", i), {n_head, n_rel_attn_bkts}, llama_model_loader::TENSOR_NOT_REQUIRED); - if (required && !found) { - throw std::runtime_error(format("key not found in model: %s", key.c_str())); - } + layer.wq = create_tensor(tn(LLM_TENSOR_DEC_ATTN_Q, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_DEC_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_DEC_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_DEC_ATTN_OUT, "weight", i), {n_embd_v_gqa, n_embd}, 0); - return found; - } + layer.attn_norm_cross = create_tensor(tn(LLM_TENSOR_DEC_CROSS_ATTN_NORM, "weight", i), {n_embd}, 0); + // this tensor seems to be unused in HF transformers implementation + layer.attn_rel_b_cross = create_tensor(tn(LLM_TENSOR_DEC_CROSS_ATTN_REL_B, "weight", i), {n_head, n_rel_attn_bkts}, llama_model_loader::TENSOR_NOT_REQUIRED); - template - bool get_key(const enum llm_kv kid, T & result, const bool required = true) { - return get_key(llm_kv(kid), result, required); - } + layer.wq_cross = create_tensor(tn(LLM_TENSOR_DEC_CROSS_ATTN_Q, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wk_cross = create_tensor(tn(LLM_TENSOR_DEC_CROSS_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wv_cross = create_tensor(tn(LLM_TENSOR_DEC_CROSS_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0); + layer.wo_cross = create_tensor(tn(LLM_TENSOR_DEC_CROSS_ATTN_OUT, "weight", i), {n_embd_v_gqa, n_embd}, 0); - // get array of n <= N_MAX elements, or a single element repeated n times - template - bool get_key_or_arr(const std::string & key, std::array & result, uint32_t n, const bool required = true) { - const int kid = gguf_find_key(meta.get(), key.c_str()); + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_DEC_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_DEC_FFN_GATE, "weight", i), {n_embd, n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_DEC_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_DEC_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_T5ENCODER: + { + const auto n_rel_attn_bkts = hparams.n_rel_attn_bkts; - if (kid < 0) { - if (required) { - throw std::runtime_error(format("key not found in model: %s", key.c_str())); - } - return false; - } + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - if (n > N_MAX) { - throw std::runtime_error(format("n > N_MAX: %u > %u for key %s", (uint32_t) n, (uint32_t) N_MAX, key.c_str())); - } + // output + model.output_norm_enc = create_tensor(tn(LLM_TENSOR_ENC_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); + // if output is NULL, init from the input tok embed + if (model.output == NULL) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); + } - if (gguf_get_kv_type(meta.get(), kid) == GGUF_TYPE_ARRAY) { - struct GGUFMeta::ArrayInfo arr_info = - GGUFMeta::GKV::get_kv(meta.get(), kid); + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - if (n != arr_info.length) { - throw std::runtime_error(format("key %s has wrong array length; expected %u, got %u", key.c_str(), n, (uint32_t) arr_info.length)); - } + layer.attn_norm_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_rel_b_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_REL_B, "weight", i), {n_head, n_rel_attn_bkts}, llama_model_loader::TENSOR_NOT_REQUIRED); - return get_arr(key, result, required); - } else { - T value; + layer.wq_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_Q, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wk_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wv_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0); + layer.wo_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_OUT, "weight", i), {n_embd_v_gqa, n_embd}, 0); - bool ok = get_key(key, value, required); - if (!ok) { - return false; - } + layer.ffn_norm_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_gate_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_GATE, "weight", i), {n_embd, n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_down_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_JAIS: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - for (uint32_t i = 0; i < n; i++) { - result[i] = value; - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - return true; - } - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - template - bool get_key_or_arr(const enum llm_kv kid, T & result, uint32_t n, const bool required = true) { - return get_key_or_arr(llm_kv(kid), result, n, required); - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); - std::string get_arch_name() const { - return arch_name; - } + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0); + layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, 0); - enum llm_arch get_arch() const { - return llm_kv.arch; - } + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0); - const llama_tensor_weight * get_weight(const char * name) const { - auto pos = weights_map.find(name); - if (pos != weights_map.end()) { - return &pos->second; - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), {n_embd}, 0); - return nullptr; - } + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, 0); - const llama_tensor_weight & require_weight(const char * name) const { - const llama_tensor_weight * weight = get_weight(name); - if (!weight) { - throw std::runtime_error(format("%s: tensor '%s' not found", __func__, name)); - } - return *weight; - } + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_gate_b = create_tensor(tn(LLM_TENSOR_FFN_GATE, "bias", i), {n_ff}, 0); - struct ggml_tensor * get_tensor_meta(const char * name) const { - const auto * weight = get_weight(name); - if (!weight) { - return nullptr; - } - return weight->tensor; - } + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, 0); + } + } break; + case LLM_ARCH_CHATGLM: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - struct ggml_tensor * require_tensor_meta(const std::string & name) const { - struct ggml_tensor * tensor = get_tensor_meta(name.c_str()); - if (!tensor) { - throw std::runtime_error(format("%s: tensor '%s' not found", __func__, name.c_str())); - } - return tensor; - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - const struct ggml_tensor * check_tensor_dims(const std::string & name, const std::vector & ne, bool required) const { - const struct ggml_tensor * cur = get_tensor_meta(name.c_str()); + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - if (cur == NULL) { - if (!required) { - return NULL; - } - throw std::runtime_error(format("%s: tensor '%s' not found", __func__, name.c_str())); - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - { - bool is_ok = true; - for (size_t i = 0; i < GGML_MAX_DIMS; ++i) { - if ((i < ne.size() && ne[i] != cur->ne[i]) || (i >= ne.size() && cur->ne[i] != 1)) { - is_ok = false; - break; - } - } - if (!is_ok) { - throw std::runtime_error( - format("%s: tensor '%s' has wrong shape; expected %s, got %s", - __func__, name.c_str(), - llama_format_tensor_shape(ne).c_str(), - llama_format_tensor_shape(cur).c_str())); - } - } + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0); + layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, 0); - return cur; - } + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - static const int TENSOR_NOT_REQUIRED = 1; - static const int TENSOR_DUPLICATED = 2; + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - struct ggml_tensor * create_tensor(struct ggml_context * ctx, const std::string & name, const std::initializer_list & ne, int flags = 0) { - const struct ggml_tensor * cur = check_tensor_dims(name, ne, !(flags & TENSOR_NOT_REQUIRED)); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff * 2}, 0); - if (cur == NULL) { - return NULL; - } + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0); + } + } break; + case LLM_ARCH_NEMOTRON: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - bool duplicated = flags & TENSOR_DUPLICATED; - - struct ggml_tensor * tensor = ggml_dup_tensor(ctx, cur); - ggml_set_name(tensor, ggml_get_name(cur)); + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - if (duplicated) { - size_data += ggml_nbytes(cur); - } else { - n_created++; - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - return tensor; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); - } + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - struct ggml_tensor * create_tensor_as_view(struct ggml_context * ctx, struct ggml_tensor * base, const std::string & name, const std::initializer_list & ne, size_t offset, bool required = true) { - const struct ggml_tensor * cur = check_tensor_dims(name, ne, required); + // optional bias tensors + layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); - if (cur == NULL) { - return NULL; - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), {n_embd}, 0); - if (cur->type != base->type) { - throw std::runtime_error(format("%s: tensor '%s' has wrong type; expected %s, got %s", __func__, name.c_str(), ggml_type_name(base->type), ggml_type_name(cur->type))); - } + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); - std::array dims; - for (size_t i = 0; i < GGML_MAX_DIMS; ++i) { - dims[i] = i < ne.size() ? ne.begin()[i] : 1; - } + // optional MLP bias + layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED); + } + } break; + case LLM_ARCH_EXAONE: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - struct ggml_tensor * tensor = ggml_view_4d(ctx, base, - dims[0], dims[1], dims[2], dims[3], - cur->nb[1], cur->nb[2], cur->nb[3], - offset); + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - ggml_set_name(tensor, name.c_str()); + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - n_created++; + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); - return tensor; - } + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_head_k * n_head}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0); - void done_getting_tensors() const { - if (n_created != n_tensors) { - throw std::runtime_error(format("%s: wrong number of tensors; expected %d, got %d", __func__, n_tensors, n_created)); - } - } + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); + layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0)); + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_RWKV6: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - void init_mappings(bool prefetch = true, llama_mlocks * mlock_mmaps = nullptr) { - if (use_mmap) { - mappings.reserve(files.size()); - mmaps_used.reserve(files.size()); - for (const auto & file : files) { - auto * reg = ggml_backend_dev_backend_reg(ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU)); - auto * is_numa_fn = (decltype(ggml_is_numa) *) ggml_backend_reg_get_proc_address(reg, "ggml_backend_cpu_is_numa"); - std::unique_ptr mapping(new llama_mmap(file.get(), prefetch ? -1 : 0, is_numa_fn())); - mmaps_used.emplace_back(mapping->size, 0); - if (mlock_mmaps) { - std::unique_ptr mlock_mmap(new llama_mlock()); - mlock_mmap->init(mapping->addr); - mlock_mmaps->emplace_back(std::move(mlock_mmap)); - } - mappings.emplace_back(std::move(mapping)); - } - } + // Block 0, LN0 + model.tok_norm = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight"), {n_embd}, 0); + model.tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias"), {n_embd}, 0); - // compute the total size of all tensors for progress reporting - for (const auto & it : weights_map) { - size_data += ggml_nbytes(it.second.tensor); - } - } + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0); - void get_mapping_range(size_t * first, size_t * last, void ** addr, int idx, ggml_context * ctx) const { - GGML_ASSERT(!mappings.empty()); - const auto & mapping = mappings.at(idx); + const int time_mix_extra_dim = hparams.time_mix_extra_dim; + const int time_decay_extra_dim = hparams.time_decay_extra_dim; + const int head_size = hparams.wkv_head_size; + const int attn_hidden_size = n_embd; + const int ffn_size = hparams.n_ff_arr[0]; - *first = mapping->size; - *last = 0; - *addr = mapping->addr; - for (ggml_tensor * tensor = ggml_get_first_tensor(ctx); tensor; tensor = ggml_get_next_tensor(ctx, tensor)) { - const auto * weight = get_weight(ggml_get_name(tensor)); - if (!weight || weight->idx != idx) { - continue; - } - *first = std::min(*first, weight->offs); - *last = std::max(*last, weight->offs + ggml_nbytes(tensor)); - } - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - // for backwards compatibility, does not support ggml-backend - void load_data_for(struct ggml_tensor * cur) const { - const auto & w = require_weight(ggml_get_name(cur)); + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0); - if (use_mmap) { - const auto & mapping = mappings.at(w.idx); - if (cur->data == nullptr) { - cur->data = (uint8_t *)mapping->addr + w.offs; - } else { - memcpy(cur->data, (uint8_t *)mapping->addr + w.offs, ggml_nbytes(cur)); - } - } else { - GGML_ASSERT(cur->data != nullptr); - GGML_ASSERT(w.idx < files.size()); - const auto & file = files.at(w.idx); - file->seek(w.offs, SEEK_SET); - file->read_raw(cur->data, ggml_nbytes(cur)); - } + layer.attn_norm_2 = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "weight", i), {n_embd}, 0); + layer.attn_norm_2_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "bias", i), {n_embd}, 0); - if (check_tensors && !ggml_validate_row_data(cur->type, cur->data, ggml_nbytes(cur))) { - throw std::runtime_error(format("tensor '%s' has invalid data", ggml_get_name(cur))); - } - } + layer.time_mix_w1 = create_tensor(tn(LLM_TENSOR_TIME_MIX_W1, "weight", i), {n_embd, time_mix_extra_dim * 5}, 0); + layer.time_mix_w2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_W2, "weight", i), {time_mix_extra_dim, n_embd, 5}, 0); - size_t size_done = 0; - size_t size_data = 0; - std::vector> mmaps_used; - - // Returns false if cancelled by progress_callback - bool load_all_data( - struct ggml_context * ctx, - llama_buf_map & bufs, - llama_mlocks * lmlocks, - llama_progress_callback progress_callback, - void * progress_callback_user_data) { - GGML_ASSERT(size_data != 0 && "call init_mappings() first"); - - std::vector> read_buf; - std::vector>> validation_result; - - // 4 staging buffers for async uploads, each sized 1MB seems to be a good default for single NVMe drives. - // NVMe raid configurations might require more / larger buffers. - constexpr size_t n_buffers = 4; - constexpr size_t buffer_size = 1 * 1024 * 1024; // 1MB - - std::vector host_buffers; - std::vector events; - std::vector host_ptrs; - size_t buffer_idx = 0; // buffer to use for async loads - ggml_backend_t upload_backend = [&](const char * func) -> ggml_backend_t { - if (use_mmap || check_tensors) { - return nullptr; - } - // When not using mmaped io use async uploads from pinned memory to GPU memory. - // First determine if the backend supports the necessary features for async uploads. - auto * buf = bufs.count(0) ? bufs.at(0) : nullptr; - if (!buf) { - LLAMA_LOG_DEBUG("%s: no buffer found for async uploads\n", func); - return nullptr; - } + layer.time_mix_lerp_x = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_X, "weight", i), {n_embd, 1, 1}, 0); + layer.time_mix_lerp_w = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_W, "weight", i), {n_embd, 1, 1}, 0); + layer.time_mix_lerp_k = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_K, "weight", i), {n_embd, 1, 1}, 0); + layer.time_mix_lerp_v = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_V, "weight", i), {n_embd, 1, 1}, 0); + layer.time_mix_lerp_r = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_R, "weight", i), {n_embd, 1, 1}, 0); + layer.time_mix_lerp_g = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_G, "weight", i), {n_embd, 1, 1}, 0); - auto * buft = ggml_backend_buffer_get_type(buf); - auto * dev = ggml_backend_buft_get_device(buft); - if (!dev) { - LLAMA_LOG_DEBUG("%s: no device found for buffer type %s for async uploads\n", func, - ggml_backend_buft_name(buft)); - return nullptr; - } + layer.time_mix_first = create_tensor(tn(LLM_TENSOR_TIME_MIX_FIRST, "weight", i), {head_size, n_embd / head_size}, 0); + layer.time_mix_decay = create_tensor(tn(LLM_TENSOR_TIME_MIX_DECAY, "weight", i), {n_embd}, 0); + layer.time_mix_decay_w1 = create_tensor(tn(LLM_TENSOR_TIME_MIX_DECAY_W1, "weight", i), {n_embd, time_decay_extra_dim}, 0); + layer.time_mix_decay_w2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_DECAY_W2, "weight", i), {time_decay_extra_dim, attn_hidden_size}, 0); + layer.time_mix_key = create_tensor(tn(LLM_TENSOR_TIME_MIX_KEY, "weight", i), {attn_hidden_size, n_embd}, 0); + layer.time_mix_value = create_tensor(tn(LLM_TENSOR_TIME_MIX_VALUE, "weight", i), {attn_hidden_size, n_embd}, 0); + layer.time_mix_receptance = create_tensor(tn(LLM_TENSOR_TIME_MIX_RECEPTANCE, "weight", i), {attn_hidden_size, n_embd}, 0); + layer.time_mix_gate = create_tensor(tn(LLM_TENSOR_TIME_MIX_GATE, "weight", i), {attn_hidden_size, n_embd}, 0); - if (buft != ggml_backend_dev_buffer_type(dev)) { - LLAMA_LOG_DEBUG("%s: buffer type %s is not the default buffer type for device %s for async uploads\n", func, - ggml_backend_buft_name(buft), ggml_backend_dev_name(dev)); - return nullptr; - } + layer.time_mix_ln = create_tensor(tn(LLM_TENSOR_TIME_MIX_LN, "weight", i), {n_embd}, 0); + layer.time_mix_ln_b = create_tensor(tn(LLM_TENSOR_TIME_MIX_LN, "bias", i), {n_embd}, 0); + layer.time_mix_output = create_tensor(tn(LLM_TENSOR_TIME_MIX_OUTPUT, "weight", i), {n_embd, attn_hidden_size}, 0); - ggml_backend_dev_props props; - ggml_backend_dev_get_props(dev, &props); - if (!props.caps.async || !props.caps.host_buffer || !props.caps.events) { - LLAMA_LOG_DEBUG("%s: device %s does not support async, host buffers or events\n", func, - ggml_backend_dev_name(dev)); - return nullptr; - } + layer.channel_mix_lerp_k = create_tensor(tn(LLM_TENSOR_CHANNEL_MIX_LERP_K, "weight", i), {n_embd, 1, 1}, 0); + layer.channel_mix_lerp_r = create_tensor(tn(LLM_TENSOR_CHANNEL_MIX_LERP_R, "weight", i), {n_embd, 1, 1}, 0); - auto * host_buft = ggml_backend_dev_host_buffer_type(dev); - if (!host_buft) { - LLAMA_LOG_DEBUG("%s: no host buffer type found for device %s\n", func, - ggml_backend_dev_name(dev)); - return nullptr; - } + layer.channel_mix_key = create_tensor(tn(LLM_TENSOR_CHANNEL_MIX_KEY, "weight", i), {n_embd, ffn_size}, 0); + layer.channel_mix_value = create_tensor(tn(LLM_TENSOR_CHANNEL_MIX_VALUE, "weight", i), {ffn_size, n_embd}, 0); + layer.channel_mix_receptance = create_tensor(tn(LLM_TENSOR_CHANNEL_MIX_RECEPTANCE, "weight", i), {n_embd, n_embd}, 0); + } - // If the backend is supported, create pinned memory buffers and events for synchronisation. - for (size_t idx = 0; idx < n_buffers; ++idx) { - auto * buf = ggml_backend_buft_alloc_buffer(host_buft, buffer_size); - if (!buf) { - LLAMA_LOG_DEBUG("%s: failed to allocate host buffer for async uploads for device %s\n", func, - ggml_backend_dev_name(dev)); - return nullptr; - } + } break; + case LLM_ARCH_CHAMELEON: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); - host_buffers.emplace_back(buf); - host_ptrs.emplace_back(ggml_backend_buffer_get_base(buf)); + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED); + // if output is NULL, init from the input tok embed + if (model.output == NULL) { + model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); + } - auto * event = ggml_backend_event_new(dev); - if (!event) { - LLAMA_LOG_DEBUG("%s: failed to create event for async uploads for device %s\n", func, - ggml_backend_dev_name(dev)); - return nullptr; - } + for (int i = 0; i < n_layer; ++i) { + auto & layer = model.layers[i]; - events.emplace_back(event); - } + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0); + layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k, n_head}, 0); + layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k, n_head_kv}, 0); + layer.attn_q_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "bias", i), {n_embd_head_k, n_head}, llama_model_loader::TENSOR_NOT_REQUIRED); + layer.attn_k_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "bias", i), {n_embd_head_k, n_head_kv}, llama_model_loader::TENSOR_NOT_REQUIRED); - ggml_backend_t backend = ggml_backend_dev_init(dev, nullptr); - if (!backend) { - LLAMA_LOG_DEBUG("%s: failed to initialize backend for device %s for async uploads\n", func, - ggml_backend_dev_name(dev)); - return nullptr; - } + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); - return backend; - }(__func__); + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0); - if (upload_backend) { - LLAMA_LOG_DEBUG("%s: using async uploads for device %s, buffer type %s, backend %s\n", __func__, - ggml_backend_dev_name(ggml_backend_get_device(upload_backend)), - ggml_backend_buft_name(ggml_backend_buffer_get_type(bufs.at(0))), - ggml_backend_name(upload_backend)); - } + layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0); + layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0); + layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0); + } + } break; + case LLM_ARCH_WAVTOKENIZER_DEC: + { + model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {hparams.n_embd_features, n_vocab}, 0); - for (struct ggml_tensor * cur = ggml_get_first_tensor(ctx); cur != NULL; cur = ggml_get_next_tensor(ctx, cur)) { - const auto * weight = get_weight(ggml_get_name(cur)); - if (weight == nullptr) { - // this can happen with split experts models - continue; - } + model.conv1d = create_tensor(tn(LLM_TENSOR_CONV1D, "weight"), {7, hparams.n_embd_features, hparams.posnet.n_embd}, 0); + model.conv1d_b = create_tensor(tn(LLM_TENSOR_CONV1D, "bias"), {1, hparams.posnet.n_embd}, 0); - if (progress_callback) { - if (!progress_callback((float) size_done / size_data, progress_callback_user_data)) { - return false; - } - } + // posnet + { + const int64_t n_embd = hparams.posnet.n_embd; + + for (uint32_t i = 0; i < hparams.posnet.n_layer; ++i) { + auto & layer = model.layers[i].posnet; + + // posnet: + // + // - resnet + // - resnet + // - attn + // - resnet + // - resnet + // - norm + // + switch (i) { + case 0: + case 1: + case 3: + case 4: + { + layer.norm1 = create_tensor(tn(LLM_TENSOR_POS_NET_NORM1, "weight", i), {1, n_embd}, 0); + layer.norm1_b = create_tensor(tn(LLM_TENSOR_POS_NET_NORM1, "bias", i), {1, n_embd}, 0); + + layer.conv1 = create_tensor(tn(LLM_TENSOR_POS_NET_CONV1, "weight", i), {3, n_embd, n_embd}, 0); + layer.conv1_b = create_tensor(tn(LLM_TENSOR_POS_NET_CONV1, "bias", i), {1, n_embd}, 0); + + layer.norm2 = create_tensor(tn(LLM_TENSOR_POS_NET_NORM2, "weight", i), {1, n_embd}, 0); + layer.norm2_b = create_tensor(tn(LLM_TENSOR_POS_NET_NORM2, "bias", i), {1, n_embd}, 0); + + layer.conv2 = create_tensor(tn(LLM_TENSOR_POS_NET_CONV2, "weight", i), {3, n_embd, n_embd}, 0); + layer.conv2_b = create_tensor(tn(LLM_TENSOR_POS_NET_CONV2, "bias", i), {1, n_embd}, 0); + } break; + case 2: + { + layer.attn_norm = create_tensor(tn(LLM_TENSOR_POS_NET_ATTN_NORM, "weight", i), {1, n_embd}, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_POS_NET_ATTN_NORM, "bias", i), {1, n_embd}, 0); + + layer.attn_q = create_tensor(tn(LLM_TENSOR_POS_NET_ATTN_Q, "weight", i), {1, n_embd, n_embd}, 0); + layer.attn_q_b = create_tensor(tn(LLM_TENSOR_POS_NET_ATTN_Q, "bias", i), {1, n_embd}, 0); + + layer.attn_k = create_tensor(tn(LLM_TENSOR_POS_NET_ATTN_K, "weight", i), {1, n_embd, n_embd}, 0); + layer.attn_k_b = create_tensor(tn(LLM_TENSOR_POS_NET_ATTN_K, "bias", i), {1, n_embd}, 0); + + layer.attn_v = create_tensor(tn(LLM_TENSOR_POS_NET_ATTN_V, "weight", i), {1, n_embd, n_embd}, 0); + layer.attn_v_b = create_tensor(tn(LLM_TENSOR_POS_NET_ATTN_V, "bias", i), {1, n_embd}, 0); + + layer.attn_o = create_tensor(tn(LLM_TENSOR_POS_NET_ATTN_OUT, "weight", i), {1, n_embd, n_embd}, 0); + layer.attn_o_b = create_tensor(tn(LLM_TENSOR_POS_NET_ATTN_OUT, "bias", i), {1, n_embd}, 0); + } break; + case 5: + { + layer.norm = create_tensor(tn(LLM_TENSOR_POS_NET_ATTN_NORM, "weight", i), {1, n_embd}, 0); + layer.norm_b = create_tensor(tn(LLM_TENSOR_POS_NET_ATTN_NORM, "bias", i), {1, n_embd}, 0); + } break; + default: GGML_ABORT("unknown posnet layer"); + }; + } + } - size_t n_size = ggml_nbytes(cur); + GGML_ASSERT(hparams.posnet.n_embd == hparams.convnext.n_embd); - if (use_mmap) { - const auto & mapping = mappings.at(weight->idx); - ggml_backend_buffer_t buf_mmap = nullptr; - if (bufs.count(weight->idx)) { - buf_mmap = bufs.at(weight->idx); - } - uint8_t * data = (uint8_t *) mapping->addr + weight->offs; + model.tok_norm = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight"), {hparams.posnet.n_embd}, 0); + model.tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias"), {hparams.posnet.n_embd}, 0); - if (check_tensors) { - validation_result.emplace_back(std::async(std::launch::async, [cur, data, n_size] { - return std::make_pair(cur, ggml_validate_row_data(cur->type, data, n_size)); - })); - } + // convnext + { + const int64_t n_embd = hparams.convnext.n_embd; - GGML_ASSERT(buf_mmap || cur->data); // either we have a buffer to allocate the tensor in, or it is already allocated - if (buf_mmap && cur->data == nullptr) { - ggml_backend_tensor_alloc(buf_mmap, cur, data); - if (lmlocks) { - const auto & lmlock = lmlocks->at(weight->idx); - lmlock->grow_to(weight->offs + n_size); - } + for (uint32_t i = 0; i < hparams.convnext.n_layer; ++i) { + auto & layer = model.layers[i].convnext; - auto & mmap_used = mmaps_used[weight->idx]; - mmap_used.first = std::min(mmap_used.first, weight->offs); - mmap_used.second = std::max(mmap_used.second, weight->offs + n_size); - } else { - ggml_backend_tensor_set(cur, data, 0, n_size); - } - } else { - const auto & file = files.at(weight->idx); - if (ggml_backend_buffer_is_host(cur->buffer)) { - file->seek(weight->offs, SEEK_SET); - file->read_raw(cur->data, n_size); - if (check_tensors) { - validation_result.emplace_back(std::async(std::launch::async, [cur, n_size] { - return std::make_pair(cur, ggml_validate_row_data(cur->type, cur->data, n_size)); - })); - } - } else { - // If upload_backend is valid load the tensor in chunks to pinned memory and upload the buffers asynchronously to the GPU. - if (upload_backend) { - file->seek(weight->offs, SEEK_SET); + layer.dw = create_tensor(tn(LLM_TENSOR_CONVNEXT_DW, "weight", i), {7, 1, n_embd}, 0); + layer.dw_b = create_tensor(tn(LLM_TENSOR_CONVNEXT_DW, "bias", i), {1, n_embd}, 0); - size_t bytes_read = 0; + layer.norm = create_tensor(tn(LLM_TENSOR_CONVNEXT_NORM, "weight", i), {n_embd}, 0); + layer.norm_b = create_tensor(tn(LLM_TENSOR_CONVNEXT_NORM, "bias", i), {n_embd}, 0); - while (bytes_read < n_size) { - size_t read_iteration = std::min(buffer_size, n_size - bytes_read); + layer.pw1 = create_tensor(tn(LLM_TENSOR_CONVNEXT_PW1, "weight", i), {n_embd, n_ff}, 0); + layer.pw1_b = create_tensor(tn(LLM_TENSOR_CONVNEXT_PW1, "bias", i), {n_ff}, 0); - ggml_backend_event_synchronize(events[buffer_idx]); - file->read_raw(host_ptrs[buffer_idx], read_iteration); - ggml_backend_tensor_set_async(upload_backend, cur, host_ptrs[buffer_idx], bytes_read, read_iteration); - ggml_backend_event_record(events[buffer_idx], upload_backend); + layer.pw2 = create_tensor(tn(LLM_TENSOR_CONVNEXT_PW2, "weight", i), {n_ff, n_embd}, 0); + layer.pw2_b = create_tensor(tn(LLM_TENSOR_CONVNEXT_PW2, "bias", i), {n_embd}, 0); - bytes_read += read_iteration; - ++buffer_idx; - buffer_idx %= n_buffers; - } - } else { - read_buf.resize(n_size); - file->seek(weight->offs, SEEK_SET); - file->read_raw(read_buf.data(), n_size); - ggml_backend_tensor_set(cur, read_buf.data(), 0, n_size); - if (check_tensors && !ggml_validate_row_data(cur->type, read_buf.data(), n_size)) { - throw std::runtime_error(format("tensor '%s' has invalid data", ggml_get_name(cur))); + layer.gamma = create_tensor(tn(LLM_TENSOR_CONVNEXT_GAMMA, "weight", i), {n_embd}, 0); } - } - } - } - - size_done += n_size; - } - - // free temporary resources used for async uploads - for (auto * event : events) { - ggml_backend_event_synchronize(event); - ggml_backend_event_free(event); - } - for (auto * buf : host_buffers) { - ggml_backend_buffer_free(buf); - } - ggml_backend_free(upload_backend); - - // check validation results - bool validation_failed = false; - for (auto & future : validation_result) { - auto result = future.get(); - if (!result.second) { - LLAMA_LOG_ERROR("%s: tensor '%s' has invalid data\n", __func__, ggml_get_name(result.first)); - validation_failed = true; - } - } - if (validation_failed) { - throw std::runtime_error("found tensors with invalid data"); - } - // check if this is the last call and do final cleanup - if (size_done >= size_data) { - // unmap offloaded tensors and metadata - if (use_mmap) { - for (uint32_t idx = 0; idx < mappings.size(); idx++) { - const auto & mmap_used = mmaps_used.at(idx); - auto & mapping = mappings.at(idx); - mapping->unmap_fragment(0, mmap_used.first); - if (mmap_used.second != 0) { - mapping->unmap_fragment(mmap_used.second, mapping->size); + // output + model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0); + model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); } - } - } - if (progress_callback) { - // Even though the model is done loading, we still honor - // cancellation since we need to free allocations. - return progress_callback(1.0f, progress_callback_user_data); - } - } - - return true; - } -}; -// temporary allocate memory for the input batch if needed -static const llama_seq_id batch_default_seq_id = 0; -struct llama_batch_allocr { - std::array seq_id_0 = {batch_default_seq_id}; - std::vector pos; - std::vector n_seq_id; - std::vector seq_id; - std::vector logits; - struct llama_batch batch; - // optionally fulfill the batch returned by llama_batch_get_one - llama_batch_allocr(llama_context & ctx, struct llama_batch in_batch) { - batch = in_batch; - GGML_ASSERT(batch.n_tokens > 0); - if (!batch.pos) { - // determine the last position in KV cache - llama_pos last_pos = -1; - for (const auto & cell : ctx.kv_self.cells) { - if (cell.has_seq_id(batch_default_seq_id)) { - last_pos = std::max(last_pos, cell.pos); - } - } - last_pos++; // next position - pos.resize(batch.n_tokens); - for (int32_t i = 0; i < batch.n_tokens; i++) { - pos[i] = i+last_pos; - } - batch.pos = pos.data(); - } - if (!batch.n_seq_id) { - n_seq_id.resize(batch.n_tokens); - for (int32_t i = 0; i < batch.n_tokens; i++) { - n_seq_id[i] = seq_id_0.size(); - } - batch.n_seq_id = n_seq_id.data(); - } - if (!batch.seq_id) { - seq_id.resize(batch.n_tokens + 1); - seq_id[batch.n_tokens] = NULL; - for (int32_t i = 0; i < batch.n_tokens; i++) { - seq_id[i] = seq_id_0.data(); - } - batch.seq_id = seq_id.data(); - } - if (!batch.logits) { - logits.resize(batch.n_tokens); - logits[logits.size() - 1] = true; - batch.logits = logits.data(); + model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {hparams.convnext.n_embd, n_embd}, 0); + model.output_b = create_tensor(tn(LLM_TENSOR_OUTPUT, "bias"), {n_embd}, 0); + } break; + default: + throw std::runtime_error("unknown architecture"); } - } -}; -template<> -bool llama_model_loader::get_key(const enum llm_kv kid, enum llama_pooling_type & result, const bool required) { - uint32_t tmp; - const bool found = get_key(kid, tmp, required); - if (found) { - result = (enum llama_pooling_type) tmp; - } else { - result = LLAMA_POOLING_TYPE_UNSPECIFIED; + if (n_moved_tensors > 0) { + LLAMA_LOG_DEBUG("%s: tensor '%s' (%s) (and %d others) cannot be used with preferred buffer type %s, using %s instead\n", + __func__, first_moved_tensor->name, ggml_type_name(first_moved_tensor->type), n_moved_tensors - 1, + ggml_backend_buft_name(first_moved_from_buft), ggml_backend_buft_name(first_moved_to_buft)); + } } - return found; -} + ml.done_getting_tensors(); -// -// load LLaMA models -// + ml.init_mappings(true, use_mlock ? &model.mlock_mmaps : nullptr); + model.mappings.reserve(ml.mappings.size()); -static const char * llama_model_arch_name(llm_arch arch) { - auto it = LLM_ARCH_NAMES.find(arch); - if (it == LLM_ARCH_NAMES.end()) { - return "unknown"; - } - return it->second; -} + // create the backend buffers + std::vector> ctx_bufs; + ctx_bufs.reserve(ctx_map.size()); -static std::string llama_model_ftype_name(llama_ftype ftype) { - if (ftype & LLAMA_FTYPE_GUESSED) { - return llama_model_ftype_name((enum llama_ftype) (ftype & ~LLAMA_FTYPE_GUESSED)) + " (guessed)"; - } + // Ensure we have enough capacity for the maximum backend buffer we will potentially create + const size_t n_max_backend_buffer = ctx_map.size() * ml.files.size(); + model.bufs.reserve(n_max_backend_buffer); - switch (ftype) { - case LLAMA_FTYPE_ALL_F32: return "all F32"; - case LLAMA_FTYPE_MOSTLY_F16: return "F16"; - case LLAMA_FTYPE_MOSTLY_BF16: return "BF16"; - case LLAMA_FTYPE_MOSTLY_Q4_0: return "Q4_0"; - case LLAMA_FTYPE_MOSTLY_Q4_1: return "Q4_1"; - case LLAMA_FTYPE_MOSTLY_Q5_0: return "Q5_0"; - case LLAMA_FTYPE_MOSTLY_Q5_1: return "Q5_1"; - case LLAMA_FTYPE_MOSTLY_Q8_0: return "Q8_0"; - case LLAMA_FTYPE_MOSTLY_Q2_K: return "Q2_K - Medium"; - case LLAMA_FTYPE_MOSTLY_Q2_K_S: return "Q2_K - Small"; - case LLAMA_FTYPE_MOSTLY_Q3_K_S: return "Q3_K - Small"; - case LLAMA_FTYPE_MOSTLY_Q3_K_M: return "Q3_K - Medium"; - case LLAMA_FTYPE_MOSTLY_Q3_K_L: return "Q3_K - Large"; - case LLAMA_FTYPE_MOSTLY_Q4_K_S: return "Q4_K - Small"; - case LLAMA_FTYPE_MOSTLY_Q4_K_M: return "Q4_K - Medium"; - case LLAMA_FTYPE_MOSTLY_Q5_K_S: return "Q5_K - Small"; - case LLAMA_FTYPE_MOSTLY_Q5_K_M: return "Q5_K - Medium"; - case LLAMA_FTYPE_MOSTLY_Q6_K: return "Q6_K"; - case LLAMA_FTYPE_MOSTLY_TQ1_0: return "TQ1_0 - 1.69 bpw ternary"; - case LLAMA_FTYPE_MOSTLY_TQ2_0: return "TQ2_0 - 2.06 bpw ternary"; - case LLAMA_FTYPE_MOSTLY_IQ2_XXS: return "IQ2_XXS - 2.0625 bpw"; - case LLAMA_FTYPE_MOSTLY_IQ2_XS: return "IQ2_XS - 2.3125 bpw"; - case LLAMA_FTYPE_MOSTLY_IQ2_S: return "IQ2_S - 2.5 bpw"; - case LLAMA_FTYPE_MOSTLY_IQ2_M: return "IQ2_M - 2.7 bpw"; - case LLAMA_FTYPE_MOSTLY_IQ3_XS: return "IQ3_XS - 3.3 bpw"; - case LLAMA_FTYPE_MOSTLY_IQ3_XXS: return "IQ3_XXS - 3.0625 bpw"; - case LLAMA_FTYPE_MOSTLY_IQ1_S: return "IQ1_S - 1.5625 bpw"; - case LLAMA_FTYPE_MOSTLY_IQ1_M: return "IQ1_M - 1.75 bpw"; - case LLAMA_FTYPE_MOSTLY_IQ4_NL: return "IQ4_NL - 4.5 bpw"; - case LLAMA_FTYPE_MOSTLY_IQ4_XS: return "IQ4_XS - 4.25 bpw"; - case LLAMA_FTYPE_MOSTLY_IQ3_S: return "IQ3_S - 3.4375 bpw"; - case LLAMA_FTYPE_MOSTLY_IQ3_M: return "IQ3_S mix - 3.66 bpw"; - - default: return "unknown, may not work"; - } -} + for (auto & it : ctx_map) { + ggml_backend_buffer_type_t buft = it.first; + ggml_context * ctx = it.second; -static const char * llama_model_type_name(e_model type) { - switch (type) { - case MODEL_14M: return "14M"; - case MODEL_17M: return "17M"; - case MODEL_22M: return "22M"; - case MODEL_33M: return "33M"; - case MODEL_60M: return "60M"; - case MODEL_70M: return "70M"; - case MODEL_80M: return "80M"; - case MODEL_109M: return "109M"; - case MODEL_137M: return "137M"; - case MODEL_160M: return "160M"; - case MODEL_220M: return "220M"; - case MODEL_250M: return "250M"; - case MODEL_270M: return "270M"; - case MODEL_335M: return "335M"; - case MODEL_410M: return "410M"; - case MODEL_450M: return "450M"; - case MODEL_770M: return "770M"; - case MODEL_780M: return "780M"; - case MODEL_0_5B: return "0.5B"; - case MODEL_1B: return "1B"; - case MODEL_1_3B: return "1.3B"; - case MODEL_1_4B: return "1.4B"; - case MODEL_1_5B: return "1.5B"; - case MODEL_1_6B: return "1.6B"; - case MODEL_2B: return "2B"; - case MODEL_2_8B: return "2.8B"; - case MODEL_3B: return "3B"; - case MODEL_4B: return "4B"; - case MODEL_6B: return "6B"; - case MODEL_6_9B: return "6.9B"; - case MODEL_7B: return "7B"; - case MODEL_8B: return "8B"; - case MODEL_9B: return "9B"; - case MODEL_11B: return "11B"; - case MODEL_12B: return "12B"; - case MODEL_13B: return "13B"; - case MODEL_14B: return "14B"; - case MODEL_15B: return "15B"; - case MODEL_16B: return "16B"; - case MODEL_20B: return "20B"; - case MODEL_30B: return "30B"; - case MODEL_32B: return "32B"; - case MODEL_34B: return "34B"; - case MODEL_35B: return "35B"; - case MODEL_40B: return "40B"; - case MODEL_65B: return "65B"; - case MODEL_70B: return "70B"; - case MODEL_236B: return "236B"; - case MODEL_314B: return "314B"; - case MODEL_SMALL: return "0.1B"; - case MODEL_MEDIUM: return "0.4B"; - case MODEL_LARGE: return "0.8B"; - case MODEL_XL: return "1.5B"; - case MODEL_A1_7B: return "A1.7B"; - case MODEL_A2_7B: return "A2.7B"; - case MODEL_8x7B: return "8x7B"; - case MODEL_8x22B: return "8x22B"; - case MODEL_16x12B: return "16x12B"; - case MODEL_10B_128x3_66B: return "10B+128x3.66B"; - case MODEL_57B_A14B: return "57B.A14B"; - case MODEL_27B: return "27B"; - default: return "?B"; - } -} + // skip contexts without tensors + if (ggml_get_first_tensor(ctx) == nullptr) { + continue; + } -static const char * llama_model_vocab_type_name(enum llama_vocab_type type){ - switch (type) { - case LLAMA_VOCAB_TYPE_NONE: return "no vocab"; - case LLAMA_VOCAB_TYPE_SPM: return "SPM"; - case LLAMA_VOCAB_TYPE_BPE: return "BPE"; - case LLAMA_VOCAB_TYPE_WPM: return "WPM"; - case LLAMA_VOCAB_TYPE_UGM: return "UGM"; - case LLAMA_VOCAB_TYPE_RWKV: return "RWKV"; - default: return "unknown"; - } -} + llama_buf_map bufs; + bufs.reserve(n_max_backend_buffer); -static void llm_load_stats(llama_model_loader & ml, llama_model & model) { - model.n_elements = ml.n_elements; - model.n_bytes = ml.n_bytes; -} + // check if it is possible to use buffer_from_host_ptr with this buffer type + ggml_backend_dev_t dev = ggml_backend_buft_get_device(buft); + if (!dev) { + // FIXME: workaround for CPU backend buft having a NULL device + dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU); + } + ggml_backend_dev_props props; + ggml_backend_dev_get_props(dev, &props); + bool buffer_from_host_ptr_supported = props.caps.buffer_from_host_ptr; + bool is_default_buft = buft == ggml_backend_dev_buffer_type(dev); -static void llm_load_arch(llama_model_loader & ml, llama_model & model) { - model.arch = ml.get_arch(); - if (model.arch == LLM_ARCH_UNKNOWN) { - throw std::runtime_error("unknown model architecture: '" + ml.get_arch_name() + "'"); - } -} + if (ml.use_mmap && use_mmap_buffer && buffer_from_host_ptr_supported && is_default_buft) { + for (uint32_t idx = 0; idx < ml.files.size(); idx++) { + // only the mmap region containing the tensors in the model is mapped to the backend buffer + // this is important for metal with apple silicon: if the entire model could be mapped to a metal buffer, then we could just use metal for all layers + // this allows using partial offloading when the model size exceeds the metal buffer size, but not the RAM size + void * addr = nullptr; + size_t first, last; // NOLINT + ml.get_mapping_range(&first, &last, &addr, idx, ctx); + if (first >= last) { + continue; + } + const size_t max_size = ggml_get_max_tensor_size(ctx); + ggml_backend_buffer_t buf = ggml_backend_dev_buffer_from_host_ptr(dev, (char *) addr + first, last - first, max_size); + if (buf == nullptr) { + throw std::runtime_error(format("unable to allocate %s buffer", ggml_backend_buft_name(buft))); + } + model.bufs.emplace_back(buf); + bufs.emplace(idx, buf); + } + } + else { + ggml_backend_buffer_t buf = ggml_backend_alloc_ctx_tensors_from_buft(ctx, buft); + if (buf == nullptr) { + throw std::runtime_error(format("unable to allocate %s buffer", ggml_backend_buft_name(buft))); + } + model.bufs.emplace_back(buf); + if (use_mlock && ggml_backend_buffer_is_host(buf)) { + model.mlock_bufs.emplace_back(new llama_mlock); + auto & mlock_buf = model.mlock_bufs.back(); + mlock_buf->init (ggml_backend_buffer_get_base(buf)); + mlock_buf->grow_to(ggml_backend_buffer_get_size(buf)); + } + for (uint32_t idx = 0; idx < ml.files.size(); idx++) { + bufs.emplace(idx, buf); + } + } -static void llm_load_hparams( - llama_model_loader & ml, - llama_model & model) { - auto & hparams = model.hparams; - const gguf_context * ctx = ml.meta.get(); + if (bufs.empty()) { + throw std::runtime_error("failed to allocate buffer"); + } - // get metadata as string - for (int i = 0; i < gguf_get_n_kv(ctx); i++) { - enum gguf_type type = gguf_get_kv_type(ctx, i); - if (type == GGUF_TYPE_ARRAY) { - continue; + for (auto & buf : bufs) { + // indicate that this buffer contains weights + // this is used by ggml_backend_sched to improve op scheduling: ops that use a weight are preferably scheduled to the backend that contains the weight + ggml_backend_buffer_set_usage(buf.second, GGML_BACKEND_BUFFER_USAGE_WEIGHTS); } - const char * name = gguf_get_key(ctx, i); - const std::string value = gguf_kv_to_str(ctx, i); - model.gguf_kv.emplace(name, value); + + ctx_bufs.emplace_back(ctx, bufs); } - // get general kv - ml.get_key(LLM_KV_GENERAL_NAME, model.name, false); + if (llama_supports_gpu_offload()) { + const int n_gpu = std::min(n_gpu_layers, int(hparams.n_layer)); + + LLAMA_LOG_INFO("%s: offloading %d repeating layers to GPU\n", __func__, n_gpu); + if (n_gpu_layers > (int) hparams.n_layer) { + LLAMA_LOG_INFO("%s: offloading output layer to GPU\n", __func__); + } - // get hparams kv - ml.get_key(LLM_KV_VOCAB_SIZE, hparams.n_vocab, false) || ml.get_arr_n(LLM_KV_TOKENIZER_LIST, hparams.n_vocab); + const int max_backend_supported_layers = hparams.n_layer + 1; + const int max_offloadable_layers = hparams.n_layer + 1; - // everything past this point is not vocab-related - if (hparams.vocab_only) { - return; + LLAMA_LOG_INFO("%s: offloaded %d/%d layers to GPU\n", __func__, std::min(n_gpu_layers, max_offloadable_layers), max_backend_supported_layers); } - ml.get_key(LLM_KV_CONTEXT_LENGTH, hparams.n_ctx_train); - ml.get_key(LLM_KV_EMBEDDING_LENGTH, hparams.n_embd); - ml.get_key(LLM_KV_BLOCK_COUNT, hparams.n_layer); - ml.get_key(LLM_KV_EXPERT_COUNT, hparams.n_expert, false); - ml.get_key(LLM_KV_EXPERT_USED_COUNT, hparams.n_expert_used, false); + // print memory requirements per buffer type + for (auto & buf : model.bufs) { + LLAMA_LOG_INFO("%s: %12s model buffer size = %8.2f MiB\n", __func__, ggml_backend_buffer_name(buf.get()), ggml_backend_buffer_get_size(buf.get()) / 1024.0 / 1024.0); + } - GGML_ASSERT(hparams.n_expert <= LLAMA_MAX_EXPERTS); - GGML_ASSERT(hparams.n_expert_used <= hparams.n_expert); - if (hparams.n_expert > 0) { - GGML_ASSERT(hparams.n_expert_used > 0); - } else { - GGML_ASSERT(hparams.n_expert_used == 0); + // populate tensors_by_name + for (auto & ctx : model.ctxs) { + for (auto * cur = ggml_get_first_tensor(ctx.get()); cur != NULL; cur = ggml_get_next_tensor(ctx.get(), cur)) { + model.tensors_by_name.emplace_back(ggml_get_name(cur), cur); + } + } + + // load tensor data + for (auto & it : ctx_bufs) { + ggml_context * ctx = it.first; + auto & bufs = it.second; + if (!ml.load_all_data(ctx, bufs, use_mlock ? &model.mlock_mmaps : NULL, progress_callback, progress_callback_user_data)) { + return false; + } + } + + if (use_mmap_buffer) { + for (auto & mapping : ml.mappings) { + model.mappings.emplace_back(std::move(mapping)); + } } - // zero-out the per-layer hparams - std::fill(hparams.n_head_arr.begin(), hparams.n_head_arr.end(), 0); - std::fill(hparams.n_head_kv_arr.begin(), hparams.n_head_kv_arr.end(), 0); - std::fill(hparams.n_ff_arr.begin(), hparams.n_ff_arr.end(), 0); + return true; +} - ml.get_key_or_arr(LLM_KV_FEED_FORWARD_LENGTH, hparams.n_ff_arr, hparams.n_layer); - ml.get_key_or_arr(LLM_KV_ATTENTION_HEAD_COUNT, hparams.n_head_arr, hparams.n_layer); +// Returns 0 on success, -1 on error, and -2 on cancellation via llama_progress_callback +static int llama_model_load(const std::string & fname, llama_model & model, llama_model_params & params) { + model.t_start_us = ggml_time_us(); - // n_head_kv is optional, default to n_head - hparams.n_head_kv_arr = hparams.n_head_arr; + try { + llama_model_loader ml(fname, params.use_mmap, params.check_tensors, params.kv_overrides); - ml.get_key_or_arr(LLM_KV_ATTENTION_HEAD_COUNT_KV, hparams.n_head_kv_arr, hparams.n_layer, false); + model.hparams.vocab_only = params.vocab_only; - bool rope_finetuned = false; - ml.get_key(LLM_KV_ROPE_SCALING_FINETUNED, rope_finetuned, false); - hparams.rope_finetuned = rope_finetuned; + try { + llm_load_arch(ml, model); + } catch(const std::exception & e) { + throw std::runtime_error("error loading model architecture: " + std::string(e.what())); + } + try { + llm_load_hparams(ml, model); + } catch(const std::exception & e) { + throw std::runtime_error("error loading model hyperparameters: " + std::string(e.what())); + } + try { + llm_load_vocab(ml, model); + } catch(const std::exception & e) { + throw std::runtime_error("error loading model vocabulary: " + std::string(e.what())); + } - hparams.n_ctx_orig_yarn = hparams.n_ctx_train; - ml.get_key(LLM_KV_ROPE_SCALING_ORIG_CTX_LEN, hparams.n_ctx_orig_yarn, false); + llm_load_stats(ml, model); + llm_load_print_meta(ml, model); - // rope_freq_base (optional) - hparams.rope_freq_base_train = 10000.0f; - ml.get_key(LLM_KV_ROPE_FREQ_BASE, hparams.rope_freq_base_train, false); + if (model.vocab.type != LLAMA_VOCAB_TYPE_NONE && + model.hparams.n_vocab != model.vocab.id_to_token.size()) { + throw std::runtime_error("vocab size mismatch"); + } - std::string rope_scaling("linear"); - ml.get_key(LLM_KV_ROPE_SCALING_TYPE, rope_scaling, false); - hparams.rope_scaling_type_train = llama_rope_scaling_type_from_string(rope_scaling); - GGML_ASSERT(hparams.rope_scaling_type_train != LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED); + if (params.vocab_only) { + LLAMA_LOG_INFO("%s: vocab only - skipping tensors\n", __func__); + return 0; + } - // rope_freq_scale (inverse of the kv) is optional - float ropescale = 0.0f; - if (!ml.get_key(LLM_KV_ROPE_SCALING_FACTOR, ropescale, false)) { - // try the old key name - ml.get_key(LLM_KV_ROPE_SCALE_LINEAR, ropescale, false); + if (!llm_load_tensors( + ml, model, params.n_gpu_layers, params.split_mode, params.main_gpu, params.tensor_split, params.use_mlock, + params.progress_callback, params.progress_callback_user_data + )) { + return -2; + } + } catch (const std::exception & err) { + LLAMA_LOG_ERROR("%s: error loading model: %s\n", __func__, err.what()); + return -1; } - hparams.rope_freq_scale_train = ropescale == 0.0f ? 1.0f : 1.0f/ropescale; - ml.get_key(LLM_KV_ROPE_SCALING_ATTN_FACTOR, hparams.rope_attn_factor, false); + // loading time will be recalculate after the first eval, so + // we take page faults deferred by mmap() into consideration + model.t_load_us = ggml_time_us() - model.t_start_us; - // non-transformer models do not have attention heads - if (hparams.n_head() > 0) { - // gpt-neox n_rot = rotary_pct * (n_embd / n_head) - // gpt-j n_rot = rotary_dim + return 0; +} - hparams.n_embd_head_k = hparams.n_embd / hparams.n_head(); - ml.get_key(LLM_KV_ATTENTION_KEY_LENGTH, hparams.n_embd_head_k, false); +// +// llm_build +// - hparams.n_embd_head_v = hparams.n_embd / hparams.n_head(); - ml.get_key(LLM_KV_ATTENTION_VALUE_LENGTH, hparams.n_embd_head_v, false); +using llm_build_cb = std::function; - // sanity check for n_rot (optional) - hparams.n_rot = hparams.n_embd_head_k; +enum llm_ffn_op_type { + LLM_FFN_SILU, + LLM_FFN_GELU, + LLM_FFN_RELU, + LLM_FFN_RELU_SQR, + LLM_FFN_SWIGLU, +}; - ml.get_key(LLM_KV_ROPE_DIMENSION_COUNT, hparams.n_rot, false); +enum llm_ffn_gate_type { + LLM_FFN_SEQ, + LLM_FFN_PAR, // ffn_gate is parallel to ffn_up +}; - if (model.arch == LLM_ARCH_LLAMA || model.arch == LLM_ARCH_FALCON) { - if (hparams.n_rot != hparams.n_embd_head_k) { - throw std::runtime_error(format("invalid n_rot: %u, expected %u", hparams.n_rot, hparams.n_embd_head_k)); - } - } +enum llm_norm_type { + LLM_NORM, + LLM_NORM_RMS, + LLM_NORM_GROUP, +}; + +static struct ggml_tensor * llm_build_inp_embd( + struct ggml_context * ctx, + struct llama_context & lctx, + const llama_hparams & hparams, + const llama_ubatch & ubatch, + struct ggml_tensor * tok_embd, + const llm_build_cb & cb) { + const int64_t n_embd = hparams.n_embd; + + struct ggml_tensor * inpL; + + if (ubatch.token) { + lctx.inp_tokens = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, ubatch.n_tokens); + cb(lctx.inp_tokens, "inp_tokens", -1); + ggml_set_input(lctx.inp_tokens); + + inpL = ggml_get_rows(ctx, tok_embd, lctx.inp_tokens); } else { - hparams.n_rot = 0; - hparams.n_embd_head_k = 0; - hparams.n_embd_head_v = 0; + lctx.inp_embd = ggml_new_tensor_2d(ctx, GGML_TYPE_F32, n_embd, ubatch.n_tokens); + inpL = lctx.inp_embd; + ggml_set_input(lctx.inp_embd); } - // arch-specific KVs - switch (model.arch) { - case LLM_ARCH_LLAMA: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); + // For Granite architecture + if (hparams.f_embedding_scale != 0.0f) { + inpL = ggml_scale(ctx, inpL, hparams.f_embedding_scale); + } - if (hparams.n_expert == 8) { - switch (hparams.n_layer) { - case 32: model.type = e_model::MODEL_8x7B; break; - case 56: model.type = e_model::MODEL_8x22B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } else { - switch (hparams.n_layer) { - case 16: model.type = e_model::MODEL_1B; break; // Llama 3.2 1B - case 22: model.type = e_model::MODEL_1B; break; - case 26: model.type = e_model::MODEL_3B; break; - case 28: model.type = e_model::MODEL_3B; break; // Llama 3.2 3B - // granite uses a vocab with len 49152 - case 32: model.type = hparams.n_vocab == 49152 ? e_model::MODEL_3B : (hparams.n_vocab < 40000 ? e_model::MODEL_7B : e_model::MODEL_8B); break; - case 36: model.type = e_model::MODEL_8B; break; // granite - case 40: model.type = e_model::MODEL_13B; break; - case 48: model.type = e_model::MODEL_34B; break; - case 60: model.type = e_model::MODEL_30B; break; - case 80: model.type = hparams.n_head() == hparams.n_head_kv() ? e_model::MODEL_65B : e_model::MODEL_70B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } - } break; - case LLM_ARCH_MINICPM: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - ml.get_key(LLM_KV_EMBEDDING_SCALE, hparams.f_embedding_scale); - ml.get_key(LLM_KV_RESIDUAL_SCALE, hparams.f_residual_scale); - ml.get_key(LLM_KV_LOGIT_SCALE, hparams.f_logit_scale); - - switch (hparams.n_layer) { - case 52: model.type = e_model::MODEL_1B; break; - case 40: model.type = e_model::MODEL_2B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_MINICPM3: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - ml.get_key(LLM_KV_ATTENTION_Q_LORA_RANK, hparams.n_lora_q); - ml.get_key(LLM_KV_ATTENTION_KV_LORA_RANK, hparams.n_lora_kv); + cb(inpL, "inp_embd", -1); - switch (hparams.n_layer) { - case 62: model.type = e_model::MODEL_4B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_GROK: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); + return inpL; +} - switch (hparams.n_layer) { - case 64: model.type = e_model::MODEL_314B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_FALCON: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); +static void llm_build_kv_store( + struct ggml_context * ctx, + const llama_hparams & hparams, + const llama_cparams & cparams, + const llama_kv_cache & kv, + struct ggml_cgraph * graph, + struct ggml_tensor * k_cur, + struct ggml_tensor * v_cur, + int32_t n_tokens, + int32_t kv_head, + const llm_build_cb & cb, + int64_t il) { + const int64_t n_ctx = cparams.n_ctx; - switch (hparams.n_layer) { - case 32: model.type = e_model::MODEL_7B; break; - case 60: model.type = e_model::MODEL_40B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_BAICHUAN: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - switch (hparams.n_layer) { - case 32: model.type = e_model::MODEL_7B; break; - case 40: model.type = e_model::MODEL_13B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } + const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(il); + const int64_t n_embd_v_gqa = hparams.n_embd_v_gqa(il); - if (model.type == e_model::MODEL_13B) { - // TODO: become GGUF KV parameter - hparams.f_max_alibi_bias = 8.0f; - } - } break; - case LLM_ARCH_STARCODER: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - switch (hparams.n_layer) { - case 24: model.type = e_model::MODEL_1B; break; - case 36: model.type = e_model::MODEL_3B; break; - case 42: model.type = e_model::MODEL_7B; break; - case 40: model.type = e_model::MODEL_15B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_REFACT: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - switch (hparams.n_layer) { - case 32: model.type = e_model::MODEL_1B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } + GGML_ASSERT(kv.size == n_ctx); - // TODO: become GGUF KV parameter - hparams.f_max_alibi_bias = 8.0f; - } break; - case LLM_ARCH_BERT: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn); - ml.get_key(LLM_KV_TOKENIZER_TOKEN_TYPE_COUNT, hparams.n_vocab_type); - ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false); - - switch (hparams.n_layer) { - case 3: - model.type = e_model::MODEL_17M; break; // bge-micro - case 6: - model.type = e_model::MODEL_22M; break; // MiniLM-L6 - case 12: - switch (hparams.n_embd) { - case 384: model.type = e_model::MODEL_33M; break; // MiniLM-L12, bge-small - case 768: model.type = e_model::MODEL_109M; break; // bge-base - } break; - case 24: - model.type = e_model::MODEL_335M; break; // bge-large - } - } break; - case LLM_ARCH_JINA_BERT_V2: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn); - ml.get_key(LLM_KV_TOKENIZER_TOKEN_TYPE_COUNT, hparams.n_vocab_type); - ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false); - hparams.f_max_alibi_bias = 8.0f; - - switch (hparams.n_layer) { - case 4: model.type = e_model::MODEL_33M; break; // jina-embeddings-small - case 12: model.type = e_model::MODEL_137M; break; // jina-embeddings-base - } - } break; - case LLM_ARCH_NOMIC_BERT: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn); - ml.get_key(LLM_KV_TOKENIZER_TOKEN_TYPE_COUNT, hparams.n_vocab_type); - ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type); + struct ggml_tensor * k_cache_view = ggml_view_1d(ctx, kv.k_l[il], n_tokens*n_embd_k_gqa, ggml_row_size(kv.k_l[il]->type, n_embd_k_gqa)*kv_head); + cb(k_cache_view, "k_cache_view", il); - if (hparams.n_layer == 12 && hparams.n_embd == 768) { - model.type = e_model::MODEL_137M; - } - } break; - case LLM_ARCH_BLOOM: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - - switch (hparams.n_layer) { - case 24: model.type = e_model::MODEL_1B; break; - case 30: - switch (hparams.n_embd) { - case 2560: model.type = e_model::MODEL_3B; break; - case 4096: model.type = e_model::MODEL_7B; break; - } break; - } + // note: storing RoPE-ed version of K in the KV cache + ggml_build_forward_expand(graph, ggml_cpy(ctx, k_cur, k_cache_view)); - // TODO: become GGUF KV parameter - hparams.f_max_alibi_bias = 8.0f; - } break; - case LLM_ARCH_MPT: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - ml.get_key(LLM_KV_ATTENTION_CLAMP_KQV, hparams.f_clamp_kqv, false); - ml.get_key(LLM_KV_ATTENTION_MAX_ALIBI_BIAS, hparams.f_max_alibi_bias); - - switch (hparams.n_layer) { - case 32: model.type = e_model::MODEL_7B; break; - case 48: model.type = e_model::MODEL_30B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_STABLELM: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - - switch (hparams.n_layer) { - case 24: model.type = e_model::MODEL_1B; break; - case 32: model.type = e_model::MODEL_3B; break; - case 40: model.type = e_model::MODEL_12B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_QWEN: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); + assert(v_cur->ne[0] == n_embd_v_gqa && v_cur->ne[1] == n_tokens); - switch (hparams.n_layer) { - case 32: model.type = e_model::MODEL_7B; break; - case 40: model.type = e_model::MODEL_13B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_QWEN2VL: - { - std::array section_dims; - ml.get_key_or_arr(LLM_KV_ROPE_DIMENSION_SECTIONS, section_dims, 4, true); - std::copy(section_dims.begin(), section_dims.begin() + 4, std::begin(hparams.rope_sections)); - } - // fall through - case LLM_ARCH_QWEN2: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - switch (hparams.n_layer) { - case 24: model.type = hparams.n_embd == 1024 ? e_model::MODEL_0_5B : e_model::MODEL_1B; break; - case 28: model.type = hparams.n_embd == 1536 ? e_model::MODEL_1_5B : e_model::MODEL_7B; break; - case 32: model.type = e_model::MODEL_7B; break; - case 36: model.type = e_model::MODEL_3B; break; - case 40: model.type = hparams.n_head() == 20 ? e_model::MODEL_4B : e_model::MODEL_13B; break; - case 48: model.type = e_model::MODEL_14B; break; - case 64: model.type = e_model::MODEL_32B; break; - case 80: model.type = e_model::MODEL_70B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_QWEN2MOE: - { - ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp, false); - ml.get_key(LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, hparams.n_ff_shexp, false); - - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - switch (hparams.n_layer) { - case 24: model.type = e_model::MODEL_A2_7B; break; - case 28: model.type = e_model::MODEL_57B_A14B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_PHI2: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); + struct ggml_tensor * v_cache_view = nullptr; - switch (hparams.n_layer) { - case 24: model.type = e_model::MODEL_1B; break; - case 32: model.type = e_model::MODEL_3B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_PHI3: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); + if (cparams.flash_attn) { + v_cache_view = ggml_view_1d(ctx, kv.v_l[il], n_tokens*n_embd_v_gqa, ggml_row_size(kv.v_l[il]->type, n_embd_v_gqa)*kv_head); + } else { + // note: the V cache is transposed when not using flash attention + v_cache_view = ggml_view_2d(ctx, kv.v_l[il], n_tokens, n_embd_v_gqa, + ( n_ctx)*ggml_element_size(kv.v_l[il]), + (kv_head)*ggml_element_size(kv.v_l[il])); - switch (hparams.n_layer) { - case 24: model.type = e_model::MODEL_1B; break; - case 32: model.type = e_model::MODEL_3B; break; - case 40: model.type = e_model::MODEL_14B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } + v_cur = ggml_transpose(ctx, v_cur); + } + cb(v_cache_view, "v_cache_view", il); - // for backward compatibility ; see: https://github.com/ggerganov/llama.cpp/pull/8931 - if ((hparams.n_layer == 32 || hparams.n_layer == 40) && hparams.n_ctx_train == 4096) { - // default value for Phi-3-mini-4k-instruct and Phi-3-medium-4k-instruct - hparams.n_swa = 2047; - } else if (hparams.n_layer == 32 && hparams.n_head_kv(0) == 32 && hparams.n_ctx_train == 131072) { - // default value for Phi-3-mini-128k-instruct - hparams.n_swa = 262144; - } else if (hparams.n_layer == 40 && hparams.n_ctx_train == 131072) { - // default value for Phi-3-medium-128k-instruct - hparams.n_swa = 131072; - } - bool found_swa = ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW, hparams.n_swa, false); - if (!found_swa && hparams.n_swa == 0) { - throw std::runtime_error("invalid value for sliding_window"); - } - } break; - case LLM_ARCH_PLAMO: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); + ggml_build_forward_expand(graph, ggml_cpy(ctx, v_cur, v_cache_view)); +} - switch (hparams.n_layer) { - case 40: model.type = e_model::MODEL_13B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_GPT2: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - switch (hparams.n_layer) { - case 12: model.type = e_model::MODEL_SMALL; break; - case 24: model.type = e_model::MODEL_MEDIUM; break; - case 36: model.type = e_model::MODEL_LARGE; break; - case 48: model.type = e_model::MODEL_XL; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_CODESHELL: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - switch (hparams.n_layer) { - case 42: model.type = e_model::MODEL_7B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_ORION: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); +// do mat_mul, while optionally apply lora +static struct ggml_tensor * llm_build_lora_mm( + struct llama_context & lctx, + struct ggml_context * ctx0, + struct ggml_tensor * w, + struct ggml_tensor * cur) { + struct ggml_tensor * res = ggml_mul_mat(ctx0, w, cur); + for (auto & it : lctx.lora_adapters) { + struct llama_lora_weight * lora = it.first->get_weight(w); + if (lora == nullptr) { + continue; + } + const float alpha = it.first->alpha; + const float rank = (float) lora->b->ne[0]; + const float scale = alpha ? it.second * alpha / rank : it.second; + struct ggml_tensor * ab_cur = ggml_mul_mat( + ctx0, lora->b, + ggml_mul_mat(ctx0, lora->a, cur) + ); + ab_cur = ggml_scale(ctx0, ab_cur, scale); + res = ggml_add(ctx0, res, ab_cur); + } + return res; +} - switch (hparams.n_layer) { - case 40: model.type = e_model::MODEL_14B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_INTERNLM2: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - switch (hparams.n_layer) { - case 32: model.type = e_model::MODEL_7B; break; - case 48: model.type = e_model::MODEL_20B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_GEMMA: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); +// do mat_mul_id, while optionally apply lora +static struct ggml_tensor * llm_build_lora_mm_id( + struct llama_context & lctx, + struct ggml_context * ctx0, + struct ggml_tensor * w, // struct ggml_tensor * as + struct ggml_tensor * cur, // struct ggml_tensor * b + struct ggml_tensor * ids) { + struct ggml_tensor * res = ggml_mul_mat_id(ctx0, w, cur, ids); + for (auto & it : lctx.lora_adapters) { + struct llama_lora_weight * lora = it.first->get_weight(w); + if (lora == nullptr) { + continue; + } + const float alpha = it.first->alpha; + const float rank = (float) lora->b->ne[0]; + const float scale = alpha ? it.second * alpha / rank : it.second; + struct ggml_tensor * ab_cur = ggml_mul_mat_id( + ctx0, lora->b, + ggml_mul_mat_id(ctx0, lora->a, cur, ids), + ids + ); + ab_cur = ggml_scale(ctx0, ab_cur, scale); + res = ggml_add(ctx0, res, ab_cur); + } + return res; +} - switch (hparams.n_layer) { - case 18: model.type = e_model::MODEL_2B; break; - case 28: model.type = e_model::MODEL_7B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_GEMMA2: +static struct ggml_tensor * llm_build_norm( + struct ggml_context * ctx, + struct ggml_tensor * cur, + const llama_hparams & hparams, + struct ggml_tensor * mw, + struct ggml_tensor * mb, + llm_norm_type type, + const llm_build_cb & cb, + int il) { + switch (type) { + case LLM_NORM: cur = ggml_norm (ctx, cur, hparams.f_norm_eps); break; + case LLM_NORM_RMS: cur = ggml_rms_norm (ctx, cur, hparams.f_norm_rms_eps); break; + case LLM_NORM_GROUP: { - hparams.n_swa = 4096; // default value of gemma 2 - ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW, hparams.n_swa, false); - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - ml.get_key(LLM_KV_ATTN_LOGIT_SOFTCAPPING, hparams.f_attn_logit_softcapping, false); - ml.get_key(LLM_KV_FINAL_LOGIT_SOFTCAPPING, hparams.f_final_logit_softcapping, false); - hparams.attn_soft_cap = true; - - switch (hparams.n_layer) { - case 26: model.type = e_model::MODEL_2B; break; - case 42: model.type = e_model::MODEL_9B; break; - case 46: model.type = e_model::MODEL_27B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } + cur = ggml_reshape_3d(ctx, cur, cur->ne[0], 1, cur->ne[1]); + cur = ggml_group_norm(ctx, cur, hparams.n_norm_groups, hparams.f_norm_group_eps); + cur = ggml_reshape_2d(ctx, cur, cur->ne[0], cur->ne[2]); } break; - case LLM_ARCH_STARCODER2: + } + + if (mw || mb) { + cb(cur, "norm", il); + } + + if (mw) { + cur = ggml_mul(ctx, cur, mw); + if (mb) { + cb(cur, "norm_w", il); + } + } + + if (mb) { + cur = ggml_add(ctx, cur, mb); + } + + return cur; +} + +static struct ggml_tensor * llm_build_ffn( + struct ggml_context * ctx, + struct llama_context & lctx, + struct ggml_tensor * cur, + struct ggml_tensor * up, + struct ggml_tensor * up_b, + struct ggml_tensor * up_s, + struct ggml_tensor * gate, + struct ggml_tensor * gate_b, + struct ggml_tensor * gate_s, + struct ggml_tensor * down, + struct ggml_tensor * down_b, + struct ggml_tensor * down_s, + struct ggml_tensor * act_scales, + llm_ffn_op_type type_op, + llm_ffn_gate_type type_gate, + const llm_build_cb & cb, + int il) { + struct ggml_tensor * tmp = up ? llm_build_lora_mm(lctx, ctx, up, cur) : cur; + cb(tmp, "ffn_up", il); + + if (up_b) { + tmp = ggml_add(ctx, tmp, up_b); + cb(tmp, "ffn_up_b", il); + } + + if (up_s) { + tmp = ggml_mul(ctx, tmp, up_s); + cb(tmp, "ffn_up_s", il); + } + + if (gate) { + switch (type_gate) { + case LLM_FFN_SEQ: + { + cur = llm_build_lora_mm(lctx, ctx, gate, tmp); + cb(cur, "ffn_gate", il); + } break; + case LLM_FFN_PAR: + { + cur = llm_build_lora_mm(lctx, ctx, gate, cur); + cb(cur, "ffn_gate", il); + } break; + } + + if (gate_b) { + cur = ggml_add(ctx, cur, gate_b); + cb(cur, "ffn_gate_b", il); + } + + if (gate_s) { + cur = ggml_mul(ctx, cur, gate_s); + cb(cur, "ffn_gate_s", il); + } + + } else { + cur = tmp; + } + + switch (type_op) { + case LLM_FFN_SILU: { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - switch (hparams.n_layer) { - case 30: model.type = e_model::MODEL_3B; break; - case 32: model.type = e_model::MODEL_7B; break; - case 40: model.type = e_model::MODEL_15B; break; - case 52: model.type = e_model::MODEL_20B; break; // granite - case 88: model.type = e_model::MODEL_34B; break; // granite - default: model.type = e_model::MODEL_UNKNOWN; - } + cur = ggml_silu(ctx, cur); + cb(cur, "ffn_silu", il); } break; - case LLM_ARCH_MAMBA: + case LLM_FFN_GELU: { - ml.get_key(LLM_KV_SSM_CONV_KERNEL, hparams.ssm_d_conv); - ml.get_key(LLM_KV_SSM_INNER_SIZE, hparams.ssm_d_inner); - ml.get_key(LLM_KV_SSM_STATE_SIZE, hparams.ssm_d_state); - ml.get_key(LLM_KV_SSM_TIME_STEP_RANK, hparams.ssm_dt_rank); - ml.get_key(LLM_KV_SSM_DT_B_C_RMS, hparams.ssm_dt_b_c_rms, false); - - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - - switch (hparams.n_layer) { - case 24: - switch (hparams.n_embd) { - case 768: model.type = e_model::MODEL_SMALL; break; - default: model.type = e_model::MODEL_UNKNOWN; - } break; - case 48: - switch (hparams.n_embd) { - case 1024: model.type = e_model::MODEL_MEDIUM; break; - case 1536: model.type = e_model::MODEL_LARGE; break; - case 2048: model.type = e_model::MODEL_XL; break; - default: model.type = e_model::MODEL_UNKNOWN; - } break; - case 64: - switch (hparams.n_embd) { - case 2560: model.type = e_model::MODEL_3B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } break; - default: model.type = e_model::MODEL_UNKNOWN; + cur = ggml_gelu(ctx, cur); + cb(cur, "ffn_gelu", il); + if (act_scales != NULL) { + cur = ggml_div(ctx, cur, act_scales); + cb(cur, "ffn_act", il); } } break; - case LLM_ARCH_XVERSE: + case LLM_FFN_RELU: { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - switch (hparams.n_layer) { - case 32: model.type = e_model::MODEL_7B; break; - case 40: model.type = e_model::MODEL_13B; break; - case 80: model.type = e_model::MODEL_65B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } + cur = ggml_relu(ctx, cur); + cb(cur, "ffn_relu", il); } break; - case LLM_ARCH_COMMAND_R: + case LLM_FFN_RELU_SQR: { - ml.get_key(LLM_KV_LOGIT_SCALE, hparams.f_logit_scale); - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - switch (hparams.n_layer) { - case 40: model.type = e_model::MODEL_35B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_DBRX: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - ml.get_key(LLM_KV_ATTENTION_CLAMP_KQV, hparams.f_clamp_kqv); + cur = ggml_relu(ctx, cur); + cb(cur, "ffn_relu", il); - switch (hparams.n_layer) { - case 40: model.type = e_model::MODEL_16x12B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_OLMO: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - ml.get_key(LLM_KV_ATTENTION_CLAMP_KQV, hparams.f_clamp_kqv, false); - - switch (hparams.n_layer) { - case 22: model.type = e_model::MODEL_1B; break; - case 32: model.type = e_model::MODEL_7B; break; - case 80: model.type = e_model::MODEL_70B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } + cur = ggml_sqr(ctx, cur); + cb(cur, "ffn_sqr(relu)", il); } break; - case LLM_ARCH_OLMO2: + case LLM_FFN_SWIGLU: { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); + // Project to 4h. If using swiglu double the output width, see https://arxiv.org/pdf/2002.05202.pdf + int64_t split_point = cur->ne[0] / 2; + struct ggml_tensor * x0 = ggml_cont(ctx, ggml_view_2d(ctx, cur, split_point, cur->ne[1], cur->nb[1], 0)); + struct ggml_tensor * x1 = ggml_cont(ctx, ggml_view_2d(ctx, cur, split_point, cur->ne[1], cur->nb[1], split_point * ggml_element_size(cur))); - switch (hparams.n_layer) { - case 16: model.type = e_model::MODEL_1B; break; - case 32: model.type = e_model::MODEL_7B; break; - case 40: model.type = e_model::MODEL_13B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_OLMOE: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - switch (hparams.n_layer) { - case 16: model.type = e_model::MODEL_A1_7B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_OPENELM: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - - switch (hparams.n_layer) { - case 16: model.type = e_model::MODEL_270M; break; - case 20: model.type = e_model::MODEL_450M; break; - case 28: model.type = e_model::MODEL_1B; break; - case 36: model.type = e_model::MODEL_3B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_GPTNEOX: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - ml.get_key(LLM_KV_USE_PARALLEL_RESIDUAL, hparams.use_par_res); - switch (hparams.n_layer) { - case 6: - switch (hparams.n_ff()) { - case 512: model.type = e_model::MODEL_14M; break; - case 2048: model.type = e_model::MODEL_70M; break; - default: model.type = e_model::MODEL_UNKNOWN; - } break; - case 12: - switch (hparams.n_ff()) { - case 3072: model.type = e_model::MODEL_160M; break; - default: model.type = e_model::MODEL_UNKNOWN; - } break; - case 16: - switch (hparams.n_ff()) { - case 8192: model.type = e_model::MODEL_1B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } break; - case 24: - switch (hparams.n_ff()) { - case 4096: model.type = e_model::MODEL_410M; break; - case 8192: model.type = e_model::MODEL_1_4B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } break; - case 32: - switch (hparams.n_ff()) { - case 10240: model.type = e_model::MODEL_2_8B; break; - case 16384: model.type = e_model::MODEL_6_9B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } break; - case 36: - switch (hparams.n_ff()) { - case 20480: model.type = e_model::MODEL_12B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } break; - case 44: - switch (hparams.n_ff()) { - case 24576: model.type = e_model::MODEL_20B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_ARCTIC: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); + x0 = ggml_silu(ctx, x0); + cb(cur, "ffn_silu", il); - if (hparams.n_expert == 128) { - switch (hparams.n_layer) { - case 35: model.type = e_model::MODEL_10B_128x3_66B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } else { - model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_DEEPSEEK: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - ml.get_key(LLM_KV_LEADING_DENSE_BLOCK_COUNT, hparams.n_layer_dense_lead); - ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp); - ml.get_key(LLM_KV_EXPERT_SHARED_COUNT, hparams.n_expert_shared); - ml.get_key(LLM_KV_EXPERT_WEIGHTS_SCALE, hparams.expert_weights_scale); - - switch (hparams.n_layer) { - case 28: model.type = e_model::MODEL_20B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_DEEPSEEK2: - { - bool is_lite = (hparams.n_layer == 27); - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - ml.get_key(LLM_KV_LEADING_DENSE_BLOCK_COUNT, hparams.n_layer_dense_lead); - if (!is_lite) { - ml.get_key(LLM_KV_ATTENTION_Q_LORA_RANK, hparams.n_lora_q); - } - ml.get_key(LLM_KV_ATTENTION_KV_LORA_RANK, hparams.n_lora_kv); - ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp); - ml.get_key(LLM_KV_EXPERT_SHARED_COUNT, hparams.n_expert_shared); - ml.get_key(LLM_KV_EXPERT_WEIGHTS_SCALE, hparams.expert_weights_scale); - ml.get_key(LLM_KV_ROPE_SCALING_YARN_LOG_MUL, hparams.rope_yarn_log_mul); - - switch (hparams.n_layer) { - case 27: model.type = e_model::MODEL_16B; break; - case 60: model.type = e_model::MODEL_236B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_CHATGLM: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - switch (hparams.n_layer) { - case 28: model.type = e_model::MODEL_6B; break; - case 40: model.type = e_model::MODEL_9B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } + cur = ggml_mul(ctx, x0, x1); + cb(cur, "ffn_mul", il); } break; - case LLM_ARCH_BITNET: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); + } - switch (hparams.n_layer) { - case 26: model.type = e_model::MODEL_3B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_T5: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - ml.get_key(LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT, hparams.n_rel_attn_bkts); + if (type_gate == LLM_FFN_PAR) { + cur = ggml_mul(ctx, cur, tmp); + cb(cur, "ffn_gate_par", il); + } - uint32_t dec_start_token_id; - if (ml.get_key(LLM_KV_DECODER_START_TOKEN_ID, dec_start_token_id, false)) { - hparams.dec_start_token_id = dec_start_token_id; - } + if (down) { + cur = llm_build_lora_mm(lctx, ctx, down, cur); + } - switch (hparams.n_layer) { - case 6: model.type = e_model::MODEL_60M; break; // t5-small - case 8: model.type = e_model::MODEL_80M; break; // flan-t5-small - case 12: - switch (hparams.n_ff()) { - case 3072: model.type = e_model::MODEL_220M; break; // t5-base - case 2048: model.type = e_model::MODEL_250M; break; // flan-t5-base - default: model.type = e_model::MODEL_UNKNOWN; - } break; - case 24: - switch (hparams.n_ff()) { - case 4096: model.type = e_model::MODEL_770M; break; // t5-large - case 2816: model.type = e_model::MODEL_780M; break; // flan-t5-large - case 16384: model.type = e_model::MODEL_3B; break; // t5-3b - case 5120: model.type = e_model::MODEL_3B; break; // flan-t5-xl - case 65536: model.type = e_model::MODEL_11B; break; // t5-11b - case 10240: model.type = e_model::MODEL_11B; break; // flan-t5-xxl - default: model.type = e_model::MODEL_UNKNOWN; - } break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_T5ENCODER: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - ml.get_key(LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT, hparams.n_rel_attn_bkts); - model.type = e_model::MODEL_UNKNOWN; - } break; - case LLM_ARCH_JAIS: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - ml.get_key(LLM_KV_ATTENTION_MAX_ALIBI_BIAS, hparams.f_max_alibi_bias); - - switch (hparams.n_layer) { - case 24: model.type = e_model::MODEL_1_3B; break; - case 40: model.type = e_model::MODEL_13B; break; - /* TODO: add variants */ - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_NEMOTRON: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - switch (hparams.n_layer) { - case 32: model.type = e_model::MODEL_4B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_EXAONE: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); + if (down_b) { + cb(cur, "ffn_down", il); + } - switch (hparams.n_layer) { - case 32: model.type = e_model::MODEL_8B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_RWKV6: - { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); - ml.get_key(LLM_KV_WKV_HEAD_SIZE, hparams.wkv_head_size); - ml.get_key(LLM_KV_TIME_MIX_EXTRA_DIM, hparams.time_mix_extra_dim); - ml.get_key(LLM_KV_TIME_DECAY_EXTRA_DIM, hparams.time_decay_extra_dim); - ml.get_key(LLM_KV_RESCALE_EVERY_N_LAYERS, hparams.rescale_every_n_layers, false); - - switch (hparams.n_layer) { - case 24: model.type = e_model::MODEL_1_6B; break; - case 32: - switch (hparams.n_embd) { - case 2560: model.type = e_model::MODEL_3B; break; - case 4096: model.type = e_model::MODEL_7B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } break; - case 61: model.type = e_model::MODEL_14B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } - } break; - case LLM_ARCH_GRANITE: - case LLM_ARCH_GRANITE_MOE: + if (down_b) { + cur = ggml_add(ctx, cur, down_b); + } + + if (down_s) { + cur = ggml_mul(ctx, cur, down_s); + cb(cur, "ffn_down_s", il); + } + + return cur; +} + +static struct ggml_tensor * llm_build_moe_ffn( + struct ggml_context * ctx, + struct llama_context & lctx, + struct ggml_tensor * cur, + struct ggml_tensor * gate_inp, + struct ggml_tensor * up_exps, + struct ggml_tensor * gate_exps, + struct ggml_tensor * down_exps, + struct ggml_tensor * exp_probs_b, + int64_t n_expert, + int64_t n_expert_used, + llm_ffn_op_type type_op, + bool norm_w, + bool scale_w, + float w_scale, +llama_expert_gating_func_type gating_op, + const llm_build_cb & cb, + int il) { + int64_t n_embd = cur->ne[0]; + int64_t n_tokens = cur->ne[1]; + + ggml_tensor * logits = llm_build_lora_mm(lctx, ctx, gate_inp, cur); // [n_expert, n_tokens] + cb(logits, "ffn_moe_logits", il); + + ggml_tensor * probs = nullptr; + switch (gating_op) { + case LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX: { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - ml.get_key(LLM_KV_LOGIT_SCALE, hparams.f_logit_scale); - ml.get_key(LLM_KV_RESIDUAL_SCALE, hparams.f_residual_scale); - ml.get_key(LLM_KV_EMBEDDING_SCALE, hparams.f_embedding_scale); - ml.get_key(LLM_KV_ATTENTION_SCALE, hparams.f_attention_scale); - - switch (hparams.n_layer) { - case 32: model.type = e_model::MODEL_3B; break; - case 40: model.type = e_model::MODEL_3B; break; - // Add additional layer/vocab/etc checks here for other model sizes - default: model.type = e_model::MODEL_UNKNOWN; - } + probs = ggml_soft_max(ctx, logits); // [n_expert, n_tokens] } break; - case LLM_ARCH_CHAMELEON: + case LLAMA_EXPERT_GATING_FUNC_TYPE_SIGMOID: { - ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); - hparams.f_norm_eps = 1e-5; // eps for qk-norm, torch default - ml.get_key(LLM_KV_SWIN_NORM, hparams.swin_norm); - - switch (hparams.n_layer) { - case 32: model.type = e_model::MODEL_7B; break; - case 48: model.type = e_model::MODEL_34B; break; - default: model.type = e_model::MODEL_UNKNOWN; - } + probs = ggml_sigmoid(ctx, logits); // [n_expert, n_tokens] } break; - default: (void)0; + default: + GGML_ABORT("fatal error"); } + cb(probs, "ffn_moe_probs", il); - model.ftype = ml.ftype; - - if (hparams.f_max_alibi_bias > 0.0f) { - hparams.use_alibi = true; + // add experts selection bias - introduced in DeepSeek V3 + // leave probs unbiased as it's later used to get expert weights + ggml_tensor * selection_probs = probs; + if (exp_probs_b != nullptr) { + selection_probs = ggml_add(ctx, probs, exp_probs_b); + cb(selection_probs, "ffn_moe_probs_biased", il); } - hparams.rope_type = llama_rope_type(&model); -} + // select experts + ggml_tensor * selected_experts = ggml_top_k(ctx, selection_probs, n_expert_used); // [n_expert_used, n_tokens] + cb(selected_experts->src[0], "ffn_moe_argsort", il); + cb(selected_experts, "ffn_moe_topk", il); -static void llm_load_vocab( - llama_model_loader & ml, - llama_model & model) { - auto & vocab = model.vocab; + ggml_tensor * weights = ggml_get_rows(ctx, + ggml_reshape_3d(ctx, probs, 1, n_expert, n_tokens), selected_experts); // [1, n_expert_used, n_tokens] + cb(weights, "ffn_moe_weights", il); - struct gguf_context * ctx = ml.meta.get(); + if (norm_w) { + weights = ggml_reshape_2d(ctx, weights, n_expert_used, n_tokens); - const auto kv = LLM_KV(model.arch); + ggml_tensor * weights_sum = ggml_sum_rows(ctx, weights); // [1, n_tokens] + cb(weights_sum, "ffn_moe_weights_sum", il); - // determine vocab type - { - std::string tokenizer_model; - std::string tokenizer_pre; - - ml.get_key(LLM_KV_TOKENIZER_MODEL, tokenizer_model); - ml.get_key(LLM_KV_TOKENIZER_PRE, tokenizer_pre, false); - - if (tokenizer_model == "no_vocab") { - vocab.type = LLAMA_VOCAB_TYPE_NONE; - - // default special tokens - vocab.special_bos_id = LLAMA_TOKEN_NULL; - vocab.special_eos_id = LLAMA_TOKEN_NULL; - vocab.special_unk_id = LLAMA_TOKEN_NULL; - vocab.special_sep_id = LLAMA_TOKEN_NULL; - vocab.special_pad_id = LLAMA_TOKEN_NULL; - vocab.special_cls_id = LLAMA_TOKEN_NULL; - vocab.special_mask_id = LLAMA_TOKEN_NULL; - vocab.linefeed_id = LLAMA_TOKEN_NULL; - - // read vocab size from metadata - if (!ml.get_key(LLM_KV_VOCAB_SIZE, vocab.n_vocab, false)) { - vocab.n_vocab = 0; - LLAMA_LOG_WARN("%s: there is no vocab_size in metadata, vocab.n_vocab will be set to %u\n", __func__, vocab.n_vocab); - } - return; - } + weights = ggml_div(ctx, weights, weights_sum); // [n_expert_used, n_tokens] + cb(weights, "ffn_moe_weights_norm", il); - if (tokenizer_model == "llama") { - vocab.type = LLAMA_VOCAB_TYPE_SPM; - - // default special tokens - vocab.special_bos_id = 1; - vocab.special_eos_id = 2; - vocab.special_unk_id = 0; - vocab.special_sep_id = LLAMA_TOKEN_NULL; - vocab.special_pad_id = LLAMA_TOKEN_NULL; - vocab.special_cls_id = LLAMA_TOKEN_NULL; - vocab.special_mask_id = LLAMA_TOKEN_NULL; - } else if (tokenizer_model == "bert") { - vocab.type = LLAMA_VOCAB_TYPE_WPM; - - // default special tokens - vocab.special_bos_id = LLAMA_TOKEN_NULL; - vocab.special_eos_id = LLAMA_TOKEN_NULL; - vocab.special_unk_id = 100; - vocab.special_sep_id = 102; - vocab.special_pad_id = 0; - vocab.special_cls_id = 101; - vocab.special_mask_id = 103; - } else if (tokenizer_model == "gpt2") { - vocab.type = LLAMA_VOCAB_TYPE_BPE; - - // read bpe merges and populate bpe ranks - const int merges_keyidx = gguf_find_key(ctx, kv(LLM_KV_TOKENIZER_MERGES).c_str()); - if (merges_keyidx == -1) { - throw std::runtime_error("cannot find tokenizer merges in model file\n"); - } + weights = ggml_reshape_3d(ctx, weights, 1, n_expert_used, n_tokens); + } + if (scale_w) { + weights = ggml_scale(ctx, weights, w_scale); + cb(weights, "ffn_moe_weights_scaled", il); + } - const int n_merges = gguf_get_arr_n(ctx, merges_keyidx); - for (int i = 0; i < n_merges; i++) { - const std::string word = gguf_get_arr_str(ctx, merges_keyidx, i); - GGML_ASSERT(unicode_cpts_from_utf8(word).size() > 0); + cur = ggml_reshape_3d(ctx, cur, n_embd, 1, n_tokens); + ggml_tensor * up = llm_build_lora_mm_id(lctx, ctx, up_exps, cur, selected_experts); // [n_ff, n_expert_used, n_tokens] + cb(up, "ffn_moe_up", il); - std::string first; - std::string second; + ggml_tensor * gate = llm_build_lora_mm_id(lctx, ctx, gate_exps, cur, selected_experts); // [n_ff, n_expert_used, n_tokens] + cb(gate, "ffn_moe_gate", il); - const size_t pos = word.find(' ', 1); + switch (type_op) { + case LLM_FFN_SILU: + { + gate = ggml_silu(ctx, gate); + cb(gate, "ffn_moe_silu", il); + } break; + case LLM_FFN_GELU: + { + gate = ggml_gelu(ctx, gate); + cb(gate, "ffn_moe_gelu", il); + } break; + default: + GGML_ABORT("fatal error"); + } - if (pos != std::string::npos) { - first = word.substr(0, pos); - second = word.substr(pos + 1); - } + ggml_tensor * par = ggml_mul(ctx, up, gate); // [n_ff, n_expert_used, n_tokens] + cb(par, "ffn_moe_gate_par", il); - vocab.bpe_ranks.emplace(std::make_pair(first, second), i); - } + ggml_tensor * experts = llm_build_lora_mm_id(lctx, ctx, down_exps, par, selected_experts); // [n_embd, n_expert_used, n_tokens] + cb(experts, "ffn_moe_down", il); - // default special tokens - vocab.special_bos_id = 11; - vocab.special_eos_id = 11; - vocab.special_unk_id = LLAMA_TOKEN_NULL; - vocab.special_sep_id = LLAMA_TOKEN_NULL; - vocab.special_pad_id = LLAMA_TOKEN_NULL; - vocab.special_cls_id = LLAMA_TOKEN_NULL; - vocab.special_mask_id = LLAMA_TOKEN_NULL; - } else if (tokenizer_model == "t5") { - vocab.type = LLAMA_VOCAB_TYPE_UGM; - - // default special tokens - vocab.special_bos_id = LLAMA_TOKEN_NULL; - vocab.special_eos_id = 1; - vocab.special_unk_id = 2; - vocab.special_sep_id = LLAMA_TOKEN_NULL; - vocab.special_pad_id = 0; - vocab.special_cls_id = LLAMA_TOKEN_NULL; - vocab.special_mask_id = LLAMA_TOKEN_NULL; - - const int precompiled_charsmap_keyidx = gguf_find_key(ctx, kv(LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP).c_str()); - if (precompiled_charsmap_keyidx != -1) { - size_t n_precompiled_charsmap = gguf_get_arr_n(ctx, precompiled_charsmap_keyidx); - const char * precompiled_charsmap = (const char *) gguf_get_arr_data(ctx, precompiled_charsmap_keyidx); - vocab.precompiled_charsmap.assign(precompiled_charsmap, precompiled_charsmap + n_precompiled_charsmap); -#ifdef IS_BIG_ENDIAN - // correct endiannes of data in precompiled_charsmap binary blob - uint32_t * xcda_blob_size = (uint32_t *) &vocab.precompiled_charsmap[0]; - *xcda_blob_size = __builtin_bswap32(*xcda_blob_size); - assert(*xcda_blob_size + sizeof(uint32_t) < n_precompiled_charsmap); - size_t xcda_array_size = *xcda_blob_size / sizeof(uint32_t); - uint32_t * xcda_array = (uint32_t *) &vocab.precompiled_charsmap[sizeof(uint32_t)]; - for (size_t i = 0; i < xcda_array_size; ++i) { - xcda_array[i] = __builtin_bswap32(xcda_array[i]); - } -#endif - } - } else if (tokenizer_model == "rwkv") { - vocab.type = LLAMA_VOCAB_TYPE_RWKV; - - // default special tokens - vocab.special_bos_id = LLAMA_TOKEN_NULL; - vocab.special_eos_id = LLAMA_TOKEN_NULL; - vocab.special_unk_id = LLAMA_TOKEN_NULL; - vocab.special_sep_id = LLAMA_TOKEN_NULL; - vocab.special_pad_id = LLAMA_TOKEN_NULL; - } else { - throw std::runtime_error(format("unknown tokenizer: '%s'", tokenizer_model.c_str())); - } + experts = ggml_mul(ctx, experts, weights); - // for now, only BPE models have pre-tokenizers - if (vocab.type == LLAMA_VOCAB_TYPE_BPE) { - vocab.tokenizer_add_space_prefix = false; - vocab.tokenizer_clean_spaces = true; - if (tokenizer_pre.empty()) { - LLAMA_LOG_WARN("%s: missing pre-tokenizer type, using: 'default'\n", __func__); - LLAMA_LOG_WARN("%s: \n", __func__); - LLAMA_LOG_WARN("%s: ************************************ \n", __func__); - LLAMA_LOG_WARN("%s: GENERATION QUALITY WILL BE DEGRADED! \n", __func__); - LLAMA_LOG_WARN("%s: CONSIDER REGENERATING THE MODEL \n", __func__); - LLAMA_LOG_WARN("%s: ************************************ \n", __func__); - LLAMA_LOG_WARN("%s: \n", __func__); - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT; - } else if (tokenizer_pre == "default") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT; - } else if ( - tokenizer_pre == "llama3" || - tokenizer_pre == "llama-v3" || - tokenizer_pre == "llama-bpe") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_LLAMA3; - vocab.tokenizer_ignore_merges = true; - vocab.tokenizer_add_bos = true; - } else if ( - tokenizer_pre == "deepseek-llm") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_LLM; - vocab.tokenizer_clean_spaces = false; - } else if ( - tokenizer_pre == "deepseek-coder") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_CODER; - vocab.tokenizer_clean_spaces = false; - } else if ( - tokenizer_pre == "falcon") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_FALCON; - } else if ( - tokenizer_pre == "falcon3") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_LLAMA3; - vocab.tokenizer_ignore_merges = true; - vocab.tokenizer_add_bos = true; - } else if ( - tokenizer_pre == "mpt") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_MPT; - } else if ( - tokenizer_pre == "starcoder") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_STARCODER; - } else if ( - tokenizer_pre == "gpt-2" || - tokenizer_pre == "phi-2" || - tokenizer_pre == "jina-es" || - tokenizer_pre == "jina-de" || - tokenizer_pre == "gigachat" || - tokenizer_pre == "jina-v1-en" || - tokenizer_pre == "jina-v2-es" || - tokenizer_pre == "jina-v2-de" || - tokenizer_pre == "jina-v2-code") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_GPT2; - } else if ( - tokenizer_pre == "refact") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_REFACT; - } else if ( - tokenizer_pre == "command-r") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_COMMAND_R; - vocab.tokenizer_clean_spaces = false; - } else if ( - tokenizer_pre == "qwen2") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_QWEN2; - vocab.tokenizer_clean_spaces = false; - } else if ( - tokenizer_pre == "stablelm2") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_STABLELM2; - } else if ( - tokenizer_pre == "olmo") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_OLMO; - } else if ( - tokenizer_pre == "dbrx") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DBRX; - } else if ( - tokenizer_pre == "smaug-bpe") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_SMAUG; - } else if ( - tokenizer_pre == "poro-chat") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_PORO; - vocab.tokenizer_clean_spaces = false; - } else if ( - tokenizer_pre == "chatglm-bpe") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_CHATGLM4; - vocab.special_bos_id = LLAMA_TOKEN_NULL; - } else if ( - tokenizer_pre == "viking") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_VIKING; - vocab.tokenizer_clean_spaces = false; - } else if ( - tokenizer_pre == "jais") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_JAIS; - } else if ( - tokenizer_pre == "tekken") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_TEKKEN; - vocab.tokenizer_clean_spaces = false; - vocab.tokenizer_ignore_merges = true; - vocab.tokenizer_add_bos = true; - } else if ( - tokenizer_pre == "smollm") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_SMOLLM; - vocab.tokenizer_clean_spaces = false; - } else if ( - tokenizer_pre == "codeshell") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_CODESHELL; - } else if ( - tokenizer_pre == "bloom") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_BLOOM; - } else if ( - tokenizer_pre == "gpt3-finnish") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISH; - } else if ( - tokenizer_pre == "exaone") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_EXAONE; - } else if ( - tokenizer_pre == "chameleon") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_CHAMELEON; - vocab.tokenizer_add_bos = true; - vocab.tokenizer_clean_spaces = false; - } else if ( - tokenizer_pre == "minerva-7b") { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_MINERVA; - } else { - throw std::runtime_error(format("unknown pre-tokenizer type: '%s'", tokenizer_pre.c_str())); - } - } else if (vocab.type == LLAMA_VOCAB_TYPE_SPM) { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT; - vocab.tokenizer_add_space_prefix = true; - vocab.tokenizer_clean_spaces = false; - vocab.tokenizer_add_bos = true; - vocab.tokenizer_add_eos = false; - } else if (vocab.type == LLAMA_VOCAB_TYPE_WPM) { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT; - vocab.tokenizer_add_space_prefix = false; - vocab.tokenizer_clean_spaces = true; - vocab.tokenizer_add_bos = true; - vocab.tokenizer_add_eos = false; - } else if (vocab.type == LLAMA_VOCAB_TYPE_UGM) { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT; - vocab.tokenizer_add_bos = false; - vocab.tokenizer_add_eos = true; - } else if (vocab.type == LLAMA_VOCAB_TYPE_RWKV) { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT; - vocab.tokenizer_add_space_prefix = false; - vocab.tokenizer_clean_spaces = false; - vocab.tokenizer_add_bos = false; - vocab.tokenizer_add_eos = false; + // aggregate experts + ggml_tensor * moe_out = nullptr; + for (int i = 0; i < n_expert_used; ++i) { + ggml_tensor * cur_expert = ggml_view_2d(ctx, experts, n_embd, n_tokens, + experts->nb[2], i*experts->nb[1]); + + if (i == 0) { + moe_out = cur_expert; } else { - vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT; + moe_out = ggml_add(ctx, moe_out, cur_expert); } - - ml.get_key(LLM_KV_TOKENIZER_ADD_PREFIX, vocab.tokenizer_add_space_prefix, false); - ml.get_key(LLM_KV_TOKENIZER_REMOVE_EXTRA_WS, vocab.tokenizer_remove_extra_whitespaces, false); } - const int token_idx = gguf_find_key(ctx, kv(LLM_KV_TOKENIZER_LIST).c_str()); - if (token_idx == -1) { - throw std::runtime_error("cannot find tokenizer vocab in model file\n"); + if (n_expert_used == 1) { + // avoid returning a non-contiguous tensor + moe_out = ggml_cont(ctx, moe_out); } - const float * scores = nullptr; - const int score_idx = gguf_find_key(ctx, kv(LLM_KV_TOKENIZER_SCORES).c_str()); - if (score_idx != -1) { - scores = (const float * ) gguf_get_arr_data(ctx, score_idx); - } + return moe_out; +} - const int * toktypes = nullptr; - const int toktype_idx = gguf_find_key(ctx, kv(LLM_KV_TOKENIZER_TOKEN_TYPE).c_str()); - if (toktype_idx != -1) { - toktypes = (const int * ) gguf_get_arr_data(ctx, toktype_idx); - } +static struct ggml_tensor * llm_build_kqv( + struct ggml_context * ctx, + struct llama_context & lctx, + const llama_kv_cache & kv, + struct ggml_cgraph * graph, + struct ggml_tensor * wo, + struct ggml_tensor * wo_b, + struct ggml_tensor * q_cur, + struct ggml_tensor * kq_mask, + int32_t n_tokens, + int32_t n_kv, + float kq_scale, + const llm_build_cb & cb, + int il) { + const llama_model & model = lctx.model; + const llama_hparams & hparams = lctx.model.hparams; + const llama_cparams & cparams = lctx.cparams; - const uint32_t n_vocab = gguf_get_arr_n(ctx, token_idx); + const int64_t n_ctx = cparams.n_ctx; + const int64_t n_head = hparams.n_head(il); + const int64_t n_head_kv = hparams.n_head_kv(il); + const int64_t n_embd_head_k = hparams.n_embd_head_k; + const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(il); + const int64_t n_embd_head_v = hparams.n_embd_head_v; + const int64_t n_embd_v_gqa = hparams.n_embd_v_gqa(il); - vocab.n_vocab = n_vocab; - vocab.id_to_token.resize(n_vocab); + struct ggml_tensor * q = ggml_permute(ctx, q_cur, 0, 2, 1, 3); + cb(q, "q", il); - for (uint32_t i = 0; i < n_vocab; i++) { - std::string word = gguf_get_arr_str(ctx, token_idx, i); + struct ggml_tensor * k = + ggml_view_3d(ctx, kv.k_l[il], + n_embd_head_k, n_kv, n_head_kv, + ggml_row_size(kv.k_l[il]->type, n_embd_k_gqa), + ggml_row_size(kv.k_l[il]->type, n_embd_head_k), + 0); + cb(k, "k", il); - //GGML_ASSERT(unicode_cpts_from_utf8(word).size() > 0); - if (word.empty()) { - LLAMA_LOG_WARN("%s: empty token at index %u\n", __func__, i); - word = "[EMPTY_" + std::to_string(i) + "]"; - } + struct ggml_tensor * cur; - vocab.token_to_id[word] = i; - vocab.max_token_len = std::max(vocab.max_token_len, (int) word.size()); - - auto & token_data = vocab.id_to_token[i]; - token_data.text = std::move(word); - token_data.score = scores ? scores[i] : 0.0f; - token_data.attr = LLAMA_TOKEN_ATTR_NORMAL; - - if (toktypes) { //TODO: remove, required until per token attributes are available from GGUF file - switch(toktypes[i]) { - case LLAMA_TOKEN_TYPE_UNKNOWN: token_data.attr = LLAMA_TOKEN_ATTR_UNKNOWN; break; - case LLAMA_TOKEN_TYPE_UNUSED: token_data.attr = LLAMA_TOKEN_ATTR_UNUSED; break; - case LLAMA_TOKEN_TYPE_NORMAL: token_data.attr = LLAMA_TOKEN_ATTR_NORMAL; break; - case LLAMA_TOKEN_TYPE_CONTROL: token_data.attr = LLAMA_TOKEN_ATTR_CONTROL; break; - case LLAMA_TOKEN_TYPE_USER_DEFINED: token_data.attr = LLAMA_TOKEN_ATTR_USER_DEFINED; break; - case LLAMA_TOKEN_TYPE_BYTE: token_data.attr = LLAMA_TOKEN_ATTR_BYTE; break; - case LLAMA_TOKEN_TYPE_UNDEFINED: token_data.attr = LLAMA_TOKEN_ATTR_UNDEFINED; break; - default: token_data.attr = LLAMA_TOKEN_ATTR_UNDEFINED; break; - } - } - } - GGML_ASSERT(vocab.id_to_token.size() == vocab.token_to_id.size()); + if (cparams.flash_attn) { + GGML_UNUSED(model); + GGML_UNUSED(n_ctx); - vocab.init_tokenizer(); + // split cached v into n_head heads (not transposed) + struct ggml_tensor * v = + ggml_view_3d(ctx, kv.v_l[il], + n_embd_head_v, n_kv, n_head_kv, + ggml_row_size(kv.v_l[il]->type, n_embd_v_gqa), + ggml_row_size(kv.v_l[il]->type, n_embd_head_v), + 0); + cb(v, "v", il); - // determine the newline token: LLaMA "<0x0A>" == 10 == '\n', Falcon 193 == '\n' - if (vocab.type == LLAMA_VOCAB_TYPE_SPM) { - try { - vocab.linefeed_id = llama_byte_to_token_impl(vocab, '\n'); - } catch (const std::exception & e) { - LLAMA_LOG_WARN("%s: SPM vocabulary, but newline token not found: %s! Using special_pad_id instead.", __func__, e.what()); - vocab.linefeed_id = vocab.special_pad_id; - } - } else if (vocab.type == LLAMA_VOCAB_TYPE_WPM) { - vocab.linefeed_id = vocab.special_pad_id; - } else if (vocab.type == LLAMA_VOCAB_TYPE_RWKV) { - const std::vector ids = llama_tokenize_internal(vocab, "\n", false); - GGML_ASSERT(!ids.empty() && "model vocab missing newline token"); - vocab.linefeed_id = ids[0]; - } else { - const std::vector ids = llama_tokenize_internal(vocab, "\xC4\x8A", false); // U+010A + cur = ggml_flash_attn_ext(ctx, q, k, v, kq_mask, kq_scale, hparams.f_max_alibi_bias, + hparams.attn_soft_cap ? hparams.f_attn_logit_softcapping : 0.0f); - //GGML_ASSERT(!ids.empty() && "model vocab missing newline token"); - if (ids.empty()) { - LLAMA_LOG_WARN("%s: model vocab missing newline token, using special_pad_id instead\n", __func__); - vocab.linefeed_id = vocab.special_pad_id; - } else { - vocab.linefeed_id = ids[0]; - } - } + ggml_flash_attn_ext_set_prec(cur, GGML_PREC_F32); - // special tokens - { - const std::vector> special_token_types = { - { LLM_KV_TOKENIZER_BOS_ID, vocab.special_bos_id }, - { LLM_KV_TOKENIZER_EOS_ID, vocab.special_eos_id }, - { LLM_KV_TOKENIZER_EOT_ID, vocab.special_eot_id }, - { LLM_KV_TOKENIZER_EOM_ID, vocab.special_eom_id }, - { LLM_KV_TOKENIZER_UNK_ID, vocab.special_unk_id }, - { LLM_KV_TOKENIZER_SEP_ID, vocab.special_sep_id }, - { LLM_KV_TOKENIZER_PAD_ID, vocab.special_pad_id }, - { LLM_KV_TOKENIZER_CLS_ID, vocab.special_cls_id }, - { LLM_KV_TOKENIZER_MASK_ID, vocab.special_mask_id }, - { LLM_KV_TOKENIZER_FIM_PRE_ID, vocab.special_fim_pre_id }, - { LLM_KV_TOKENIZER_FIM_SUF_ID, vocab.special_fim_suf_id }, - { LLM_KV_TOKENIZER_FIM_MID_ID, vocab.special_fim_mid_id }, - { LLM_KV_TOKENIZER_FIM_PAD_ID, vocab.special_fim_pad_id }, - { LLM_KV_TOKENIZER_FIM_REP_ID, vocab.special_fim_rep_id }, - { LLM_KV_TOKENIZER_FIM_SEP_ID, vocab.special_fim_sep_id }, - - // deprecated - { LLM_KV_TOKENIZER_PREFIX_ID, vocab.special_fim_pre_id }, - { LLM_KV_TOKENIZER_SUFFIX_ID, vocab.special_fim_suf_id }, - { LLM_KV_TOKENIZER_MIDDLE_ID, vocab.special_fim_mid_id }, - }; + cur = ggml_reshape_2d(ctx, cur, n_embd_head_v*n_head, n_tokens); + } else { + struct ggml_tensor * kq = ggml_mul_mat(ctx, k, q); + cb(kq, "kq", il); - for (const auto & it : special_token_types) { - const std::string & key = kv(std::get<0>(it)); - int32_t & id = std::get<1>(it); + // note: this op tends to require high floating point range + // while for some models F16 is enough, for others it is not, so we default to F32 here + ggml_mul_mat_set_prec(kq, GGML_PREC_F32); - uint32_t new_id; - if (!ml.get_key(std::get<0>(it), new_id, false)) { - continue; - } - if (new_id >= vocab.id_to_token.size()) { - LLAMA_LOG_WARN("%s: bad special token: '%s' = %ud, using default id %d\n", - __func__, key.c_str(), new_id, id); - } else { - id = new_id; - } - } - - // Handle add_bos_token and add_eos_token - { - bool temp = true; - - if (ml.get_key(LLM_KV_TOKENIZER_ADD_BOS, temp, false)) { - vocab.tokenizer_add_bos = temp; - } - if (ml.get_key(LLM_KV_TOKENIZER_ADD_EOS, temp, false)) { - vocab.tokenizer_add_eos = temp; - } - } - - // auto-detect special tokens by text - // TODO: convert scripts should provide these tokens through the KV metadata LLM_KV_TOKENIZER_... - // for now, we apply this workaround to find the tokens based on their text - - for (const auto & t : vocab.token_to_id) { - // find EOT token: "<|eot_id|>", "<|im_end|>", "", etc. - if (vocab.special_eot_id == LLAMA_TOKEN_NULL) { - if (false - || t.first == "<|eot_id|>" - || t.first == "<|im_end|>" - || t.first == "<|end|>" - || t.first == "" - || t.first == "<|endoftext|>" - || t.first == "" - || t.first == "<｜end▁of▁sentence｜>" // DeepSeek - ) { - vocab.special_eot_id = t.second; - if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) { - LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n", - __func__, t.second, t.first.c_str()); - vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL; - } - } - } - - // find EOM token: "<|eom_id|>" - if (vocab.special_eom_id == LLAMA_TOKEN_NULL) { - if (false - || t.first == "<|eom_id|>" - ) { - vocab.special_eom_id = t.second; - if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) { - LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n", - __func__, t.second, t.first.c_str()); - vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL; - } - } - } - - // find FIM_PRE token: "<|fim_prefix|>", "", "
", etc.
-            if (vocab.special_fim_pre_id == LLAMA_TOKEN_NULL) {
-                if (false
-                        || t.first == "<|fim_prefix|>"  // Qwen
-                        || t.first == ""
-                        || t.first == "<｜fim▁begin｜>" // DeepSeek
-                        || t.first == "
"
-                        ) {
-                    vocab.special_fim_pre_id = t.second;
-                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.second, t.first.c_str());
-                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
-                    }
-                }
-            }
-
-            // find FIM_SUF token: "<|fim_suffix|>", "", "", etc.
-            if (vocab.special_fim_suf_id == LLAMA_TOKEN_NULL) {
-                if (false
-                        || t.first == "<|fim_suffix|>" // Qwen
-                        || t.first == ""
-                        || t.first == "<｜fim▁hole｜>" // DeepSeek
-                        || t.first == ""
-                        ) {
-                    vocab.special_fim_suf_id = t.second;
-                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.second, t.first.c_str());
-                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
-                    }
-                }
-            }
-
-            // find FIM_MID token: "<|fim_middle|>", "", "", etc.
-            if (vocab.special_fim_mid_id == LLAMA_TOKEN_NULL) {
-                if (false
-                        || t.first == "<|fim_middle|>" // Qwen
-                        || t.first == ""
-                        || t.first == "<｜fim▁end｜>"  // DeepSeek
-                        || t.first == ""
-                        ) {
-                    vocab.special_fim_mid_id = t.second;
-                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.second, t.first.c_str());
-                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
-                    }
-                }
-            }
-
-            // find FIM_PAD token: "<|fim_pad|>", "", "", etc.
-            if (vocab.special_fim_pad_id == LLAMA_TOKEN_NULL) {
-                if (false
-                        || t.first == "<|fim_pad|>" // Qwen
-                        || t.first == ""
-                        || t.first == ""
-                        ) {
-                    vocab.special_fim_pad_id = t.second;
-                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.second, t.first.c_str());
-                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
-                    }
-                }
-            }
-
-            // find FIM_REP token: "<|fim_repo|>", "", "", etc.
-            if (vocab.special_fim_rep_id == LLAMA_TOKEN_NULL) {
-                if (false
-                        || t.first == "<|fim_repo|>"  // Qwen
-                        || t.first == "<|repo_name|>"
-                        || t.first == ""
-                        || t.first == ""
-                        ) {
-                    vocab.special_fim_rep_id = t.second;
-                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.second, t.first.c_str());
-                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
-                    }
-                }
-            }
+        if (model.arch == LLM_ARCH_GROK) {
+            // need to do the following:
+            // multiply by attn_output_multiplyer of 0.08838834764831845
+            // and then :
+            // kq = 30 * tanh(kq / 30)
+            // before the softmax below
 
-            // find FIM_SEP token: "<|file_sep|>"
-            if (vocab.special_fim_sep_id == LLAMA_TOKEN_NULL) {
-                if (false
-                        || t.first == "<|file_sep|>" // Qwen
-                        ) {
-                    vocab.special_fim_sep_id = t.second;
-                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.second, t.first.c_str());
-                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
-                    }
-                }
-            }
+            kq = ggml_tanh(ctx, ggml_scale(ctx, kq, 0.08838834764831845f/30.0f));
+            kq = ggml_scale(ctx, kq, 30);
         }
 
-        // maintain a list of tokens that cause end-of-generation
-        // this is currently determined based on the token text, which is obviously not ideal
-        // ref: https://github.com/ggerganov/llama.cpp/issues/9606
-        vocab.special_eog_ids.clear();
-
-        if (vocab.special_fim_pad_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_fim_pad_id) == 0) {
-            vocab.special_eog_ids.insert(vocab.special_fim_pad_id);
+        if (hparams.attn_soft_cap) {
+            kq = ggml_scale(ctx, kq, 1.0f / hparams.f_attn_logit_softcapping);
+            kq = ggml_tanh(ctx, kq);
+            kq = ggml_scale(ctx, kq, hparams.f_attn_logit_softcapping);
         }
 
-        if (vocab.special_fim_rep_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_fim_rep_id) == 0) {
-            vocab.special_eog_ids.insert(vocab.special_fim_rep_id);
-        }
+        kq = ggml_soft_max_ext(ctx, kq, kq_mask, kq_scale, hparams.f_max_alibi_bias);
+        cb(kq, "kq_soft_max_ext", il);
 
-        if (vocab.special_fim_sep_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_fim_sep_id) == 0) {
-            vocab.special_eog_ids.insert(vocab.special_fim_sep_id);
-        }
+        GGML_ASSERT(kv.size == n_ctx);
 
-        for (const auto & t : vocab.token_to_id) {
-            if (false
-                    || t.first == "<|eot_id|>"
-                    || t.first == "<|im_end|>"
-                    || t.first == "<|end|>"
-                    || t.first == ""
-                    || t.first == "<|endoftext|>"
-                    || t.first == "<|eom_id|>"
-                    || t.first == ""
-               ) {
-                vocab.special_eog_ids.insert(t.second);
-                if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                    LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                            __func__, t.second, t.first.c_str());
-                    vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
-                }
-            } else {
-                // token is control, but not marked as EOG -> print a debug log
-                if (vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL && vocab.special_eog_ids.count(t.second) == 0) {
-                    LLAMA_LOG_DEBUG("%s: control token: %6d '%s' is not marked as EOG\n",
-                            __func__, t.second, t.first.c_str());
-                }
-            }
-        }
+        // split cached v into n_head heads
+        struct ggml_tensor * v =
+            ggml_view_3d(ctx, kv.v_l[il],
+                    n_kv, n_embd_head_v, n_head_kv,
+                    ggml_element_size(kv.v_l[il])*n_ctx,
+                    ggml_element_size(kv.v_l[il])*n_ctx*n_embd_head_v,
+                    0);
+        cb(v, "v", il);
 
-        // sanity checks
-        if (vocab.special_eos_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_eos_id) == 0) {
-            vocab.special_eog_ids.insert(vocab.special_eos_id);
-            LLAMA_LOG_WARN("%s: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect\n", __func__);
-        }
+        struct ggml_tensor * kqv = ggml_mul_mat(ctx, v, kq);
+        cb(kqv, "kqv", il);
 
-        if (vocab.special_eot_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_eot_id) == 0) {
-            vocab.special_eog_ids.insert(vocab.special_eot_id);
-            LLAMA_LOG_WARN("%s: special_eot_id is not in special_eog_ids - the tokenizer config may be incorrect\n", __func__);
-        }
+        struct ggml_tensor * kqv_merged = ggml_permute(ctx, kqv, 0, 2, 1, 3);
+        cb(kqv_merged, "kqv_merged", il);
 
-        if (vocab.special_eom_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_eom_id) == 0) {
-            vocab.special_eog_ids.insert(vocab.special_eom_id);
-            LLAMA_LOG_WARN("%s: special_eom_id is not in special_eog_ids - the tokenizer config may be incorrect\n", __func__);
-        }
+        cur = ggml_cont_2d(ctx, kqv_merged, n_embd_head_v*n_head, n_tokens);
+        cb(cur, "kqv_merged_cont", il);
     }
 
-    // build special tokens cache
-    {
-        for (llama_vocab::id id = 0; id < (llama_vocab::id)n_vocab; ++id) {
-            if (vocab.id_to_token[id].attr & (LLAMA_TOKEN_ATTR_CONTROL | LLAMA_TOKEN_ATTR_USER_DEFINED | LLAMA_TOKEN_ATTR_UNKNOWN)) {
-                vocab.cache_special_tokens.push_back(id);
-            }
-        }
+    ggml_build_forward_expand(graph, cur);
 
-        std::sort(vocab.cache_special_tokens.begin(), vocab.cache_special_tokens.end(),
-            [&] (const llama_vocab::id a, const llama_vocab::id b) {
-                return vocab.id_to_token[a].text.size() > vocab.id_to_token[b].text.size();
-            }
-        );
+    if (wo) {
+        cur = llm_build_lora_mm(lctx, ctx, wo, cur);
+    }
 
-        LLAMA_LOG_INFO("%s: special tokens cache size = %u\n", __func__, (uint32_t)vocab.cache_special_tokens.size());
+    if (wo_b) {
+        cb(cur, "kqv_wo", il);
     }
 
-    // build token to piece cache
-    {
-        size_t size_cache = 0;
+    if (wo_b) {
+        cur = ggml_add(ctx, cur, wo_b);
+    }
 
-        std::vector cache_token_to_piece(n_vocab);
+    return cur;
+}
 
-        for (uint32_t id = 0; id < n_vocab; ++id) {
-            cache_token_to_piece[id] = llama_token_to_piece(&model, id, true);
+static struct ggml_tensor * llm_build_kv(
+        struct ggml_context * ctx,
+       struct llama_context & lctx,
+       const llama_kv_cache & kv,
+         struct ggml_cgraph * graph,
+         struct ggml_tensor * wo,
+         struct ggml_tensor * wo_b,
+         struct ggml_tensor * k_cur,
+         struct ggml_tensor * v_cur,
+         struct ggml_tensor * q_cur,
+         struct ggml_tensor * kq_mask,
+                    int32_t   n_tokens,
+                    int32_t   kv_head,
+                    int32_t   n_kv,
+                    float     kq_scale,
+         const llm_build_cb & cb,
+                    int       il) {
+    const llama_hparams & hparams = lctx.model.hparams;
+    const llama_cparams & cparams = lctx.cparams;
 
-            size_cache += cache_token_to_piece[id].size();
-        }
+    // these nodes are added to the graph together so that they are not reordered
+    // by doing so, the number of splits in the graph is reduced
+    ggml_build_forward_expand(graph, q_cur);
+    ggml_build_forward_expand(graph, k_cur);
+    ggml_build_forward_expand(graph, v_cur);
 
-        std::swap(vocab.cache_token_to_piece, cache_token_to_piece);
+    llm_build_kv_store(ctx, hparams, cparams, kv, graph, k_cur, v_cur, n_tokens, kv_head, cb, il);
 
-        LLAMA_LOG_INFO("%s: token to piece cache size = %.4f MB\n", __func__, size_cache / 1024.0 / 1024.0);
-    }
+    struct ggml_tensor * cur;
 
-    // Handle per token attributes
-    //NOTE: Each model customizes per token attributes.
-    //NOTE: Per token attributes are missing from the GGUF file.
-    //TODO: Extract attributes from GGUF file.
-    {
-        auto _contains_any = [] (const std::string &str, const std::vector &substrs) -> bool {
-            for (auto substr : substrs) {
-                if (str.find(substr) < std::string::npos) {
-                    return true;
-                }
-            }
-            return false;
-        };
+    cur  = llm_build_kqv(ctx, lctx, kv, graph, wo, wo_b, q_cur, kq_mask, n_tokens, n_kv, kq_scale, cb, il);
+    cb(cur, "kqv_out", il);
 
-        auto _set_tokenid_attr = [&] (const llama_vocab::id id, llama_token_attr attr, bool value) {
-            uint32_t current = vocab.id_to_token.at(id).attr;
-            current = value ? (current | attr) : (current & ~attr);
-            vocab.id_to_token[id].attr = (llama_token_attr) current;
-        };
+    return cur;
+}
 
-        auto _set_token_attr = [&] (const std::string & token, llama_token_attr attr, bool value) {
-            _set_tokenid_attr(vocab.token_to_id.at(token), attr, value);
-        };
+static struct ggml_tensor * llm_build_copy_mask_state(
+        struct ggml_context * ctx,
+         struct ggml_cgraph * graph,
+         struct ggml_tensor * s,
+         struct ggml_tensor * state_copy,
+         struct ggml_tensor * state_mask,
+                    int32_t   n_state,
+                    int32_t   kv_size,
+                    int32_t   kv_head,
+                    int32_t   n_kv,
+                    int32_t   n_seqs) {
+    struct ggml_tensor * states = ggml_reshape_2d(ctx, s, n_state, kv_size);
 
-        std::string model_name;
-        std::string tokenizer_pre;
+    // copy states
+    // NOTE: assuming the copy destinations are ALL contained between kv_head and kv_head + n_kv
+    // this shrinks the tensors's ne[1] to n_kv
+    states = ggml_get_rows(ctx, states, state_copy);
 
-        ml.get_key(LLM_KV_GENERAL_NAME, model_name, false);
-        ml.get_key(LLM_KV_TOKENIZER_PRE, tokenizer_pre, false);
+    // clear states of sequences which are starting at the beginning of this batch
+    // FIXME: zero-out NANs?
+    states = ggml_mul(ctx, states, state_mask);
 
-        // model name to lowercase
-        std::transform(model_name.begin(), model_name.end(), model_name.begin(),
-            [] (const std::string::value_type x) {
-                return std::tolower(x);
-            }
-        );
+    // copy states which won't be changed further (between n_seqs and n_kv)
+    ggml_build_forward_expand(graph,
+        ggml_cpy(ctx,
+            ggml_view_1d(ctx, states, n_state*(n_kv - n_seqs), n_seqs*n_state*ggml_element_size(states)),
+            ggml_view_1d(ctx, s, n_state*(n_kv - n_seqs), (kv_head + n_seqs)*n_state*ggml_element_size(s))));
 
-        // set attributes by model/tokenizer name
-        if (_contains_any(tokenizer_pre, {"jina-v2-de", "jina-v2-es", "jina-v2-code"})) {
-            _set_token_attr("", LLAMA_TOKEN_ATTR_LSTRIP, true);
-        } else if (_contains_any(model_name, {"phi-3", "phi3"})) {
-            for (auto id : vocab.cache_special_tokens) {
-                _set_tokenid_attr(id, LLAMA_TOKEN_ATTR_RSTRIP, true);
-            }
-            for (auto token : {""}) {
-                _set_token_attr(token, LLAMA_TOKEN_ATTR_RSTRIP, true);
-            }
-            for (auto token : {"", "", "<|endoftext|>"}) {
-                _set_token_attr(token, LLAMA_TOKEN_ATTR_RSTRIP, false);
-            }
-        }
-    }
+    // the part of the states that will be used and modified
+    return ggml_view_2d(ctx, states, n_state, n_seqs, states->nb[1], 0);
 }
 
-static void llm_load_print_meta(llama_model_loader & ml, llama_model & model) {
-    const auto & hparams = model.hparams;
-    const auto & vocab   = model.vocab;
-
-    const char * rope_scaling_type = LLAMA_ROPE_SCALING_TYPES.at(hparams.rope_scaling_type_train);
-
-    auto print_f = [](const std::function & f, uint32_t n) {
-        bool is_var = false;
+// TODO: split
+static struct ggml_tensor * llm_build_mamba(
+        struct ggml_context * ctx,
+       struct llama_context & lctx,
+         const llama_ubatch & ubatch,
+         struct ggml_cgraph * graph,
+         struct ggml_tensor * cur,
+         struct ggml_tensor * state_copy,
+         struct ggml_tensor * state_mask,
+                    int32_t   kv_head,
+                    int32_t   n_kv,
+         const llm_build_cb & cb,
+                    int       il) {
+    const llama_model    & model   = lctx.model;
+    const llama_hparams  & hparams = model.hparams;
+    const llama_kv_cache & kv      = lctx.kv_self;
+    const int64_t d_conv  = hparams.ssm_d_conv;
+    const int64_t d_inner = hparams.ssm_d_inner;
+    const int64_t d_state = hparams.ssm_d_state;
+    const int64_t dt_rank = hparams.ssm_dt_rank;
+    const int64_t n_seqs  = ubatch.n_seqs;
+    // Some variants of Mamba arch (e.g. FalconMamba do apply layer norm on B and Dt layers)
+    const bool ssm_dt_b_c_rms = hparams.ssm_dt_b_c_rms;
+    // Use the same RMS norm as the final layer norm
+    const float norm_rms_eps = hparams.f_norm_rms_eps;
 
-        std::vector v;
-        for (uint32_t i = 0; i < n; ++i) {
-            v.push_back(f(i));
-            if (v[i] != v[0]) {
-                is_var = true;
-            }
-        }
+    const int64_t n_seq_tokens = ubatch.n_seq_tokens;
 
-        std::stringstream ss;
+    GGML_ASSERT(n_seqs != 0);
+    GGML_ASSERT(ubatch.equal_seqs);
+    GGML_ASSERT(ubatch.n_tokens == n_seq_tokens * n_seqs);
 
-        if (is_var) {
-            ss << "[";
-            for (uint32_t i = 0; i < n; ++i) {
-                ss << v[i];
-                if (i < n - 1) {
-                    ss << ", ";
-                }
-            }
-            ss << "]";
-        } else {
-            ss << v[0];
-        }
+    struct ggml_tensor * conv_states_all = kv.k_l[il];
+    struct ggml_tensor * ssm_states_all  = kv.v_l[il];
 
-        return ss.str();
-    };
+    // (ab)using the KV cache to store the states
+    struct ggml_tensor * conv = llm_build_copy_mask_state(ctx,
+            graph, conv_states_all, state_copy, state_mask,
+            hparams.n_embd_k_s(), kv.size, kv_head, n_kv, n_seqs);
+    conv = ggml_reshape_3d(ctx, conv, d_conv - 1, d_inner, n_seqs);
+    struct ggml_tensor * ssm = llm_build_copy_mask_state(ctx,
+            graph, ssm_states_all, state_copy, state_mask,
+            hparams.n_embd_v_s(), kv.size, kv_head, n_kv, n_seqs);
+    ssm = ggml_reshape_3d(ctx, ssm, d_state, d_inner, n_seqs);
 
-    // hparams
-    LLAMA_LOG_INFO("%s: format           = %s\n",     __func__, llama_file_version_name(ml.fver));
-    LLAMA_LOG_INFO("%s: arch             = %s\n",     __func__, LLM_ARCH_NAMES.at(model.arch));
-    LLAMA_LOG_INFO("%s: vocab type       = %s\n",     __func__, llama_model_vocab_type_name(vocab.type));
-    LLAMA_LOG_INFO("%s: n_vocab          = %u\n",     __func__, hparams.n_vocab);
-    LLAMA_LOG_INFO("%s: n_merges         = %u\n",     __func__, (int) vocab.bpe_ranks.size());
-    LLAMA_LOG_INFO("%s: vocab_only       = %d\n",     __func__, hparams.vocab_only);
+    // {n_embd, n_tokens} => {n_embd, n_seq_tokens, n_seqs}
+    cur = ggml_reshape_3d(ctx, cur, cur->ne[0], n_seq_tokens, n_seqs);
 
-    if (!hparams.vocab_only) {
-        LLAMA_LOG_INFO("%s: n_ctx_train      = %u\n",     __func__, hparams.n_ctx_train);
-        LLAMA_LOG_INFO("%s: n_embd           = %u\n",     __func__, hparams.n_embd);
-        LLAMA_LOG_INFO("%s: n_layer          = %u\n",     __func__, hparams.n_layer);
-        LLAMA_LOG_INFO("%s: n_head           = %s\n",     __func__, print_f([&](uint32_t il) { return hparams.n_head(il);    }, hparams.n_layer).c_str());
-        LLAMA_LOG_INFO("%s: n_head_kv        = %s\n",     __func__, print_f([&](uint32_t il) { return hparams.n_head_kv(il); }, hparams.n_layer).c_str());
-        LLAMA_LOG_INFO("%s: n_rot            = %u\n",     __func__, hparams.n_rot);
-        LLAMA_LOG_INFO("%s: n_swa            = %u\n",     __func__, hparams.n_swa);
-        LLAMA_LOG_INFO("%s: n_embd_head_k    = %u\n",     __func__, hparams.n_embd_head_k);
-        LLAMA_LOG_INFO("%s: n_embd_head_v    = %u\n",     __func__, hparams.n_embd_head_v);
-        LLAMA_LOG_INFO("%s: n_gqa            = %s\n",     __func__, print_f([&](uint32_t il) { return hparams.n_gqa(il);        }, hparams.n_layer).c_str());
-        LLAMA_LOG_INFO("%s: n_embd_k_gqa     = %s\n",     __func__, print_f([&](uint32_t il) { return hparams.n_embd_k_gqa(il); }, hparams.n_layer).c_str());
-        LLAMA_LOG_INFO("%s: n_embd_v_gqa     = %s\n",     __func__, print_f([&](uint32_t il) { return hparams.n_embd_v_gqa(il); }, hparams.n_layer).c_str());
-        LLAMA_LOG_INFO("%s: f_norm_eps       = %.1e\n",   __func__, hparams.f_norm_eps);
-        LLAMA_LOG_INFO("%s: f_norm_rms_eps   = %.1e\n",   __func__, hparams.f_norm_rms_eps);
-        LLAMA_LOG_INFO("%s: f_clamp_kqv      = %.1e\n",   __func__, hparams.f_clamp_kqv);
-        LLAMA_LOG_INFO("%s: f_max_alibi_bias = %.1e\n",   __func__, hparams.f_max_alibi_bias);
-        LLAMA_LOG_INFO("%s: f_logit_scale    = %.1e\n",   __func__, hparams.f_logit_scale);
-        LLAMA_LOG_INFO("%s: n_ff             = %s\n",     __func__, print_f([&](uint32_t il) { return hparams.n_ff(il); }, hparams.n_layer).c_str());
-        LLAMA_LOG_INFO("%s: n_expert         = %u\n",     __func__, hparams.n_expert);
-        LLAMA_LOG_INFO("%s: n_expert_used    = %u\n",     __func__, hparams.n_expert_used);
-        LLAMA_LOG_INFO("%s: causal attn      = %d\n",     __func__, hparams.causal_attn);
-        LLAMA_LOG_INFO("%s: pooling type     = %d\n",     __func__, hparams.pooling_type);
-        LLAMA_LOG_INFO("%s: rope type        = %d\n",     __func__, hparams.rope_type);
-        LLAMA_LOG_INFO("%s: rope scaling     = %s\n",     __func__, rope_scaling_type);
-        LLAMA_LOG_INFO("%s: freq_base_train  = %.1f\n",   __func__, hparams.rope_freq_base_train);
-        LLAMA_LOG_INFO("%s: freq_scale_train = %g\n",     __func__, hparams.rope_freq_scale_train);
-        LLAMA_LOG_INFO("%s: n_ctx_orig_yarn  = %u\n",     __func__, hparams.n_ctx_orig_yarn);
-        LLAMA_LOG_INFO("%s: rope_finetuned   = %s\n",     __func__, hparams.rope_finetuned ? "yes" : "unknown");
-        LLAMA_LOG_INFO("%s: ssm_d_conv       = %u\n",     __func__, hparams.ssm_d_conv);
-        LLAMA_LOG_INFO("%s: ssm_d_inner      = %u\n",     __func__, hparams.ssm_d_inner);
-        LLAMA_LOG_INFO("%s: ssm_d_state      = %u\n",     __func__, hparams.ssm_d_state);
-        LLAMA_LOG_INFO("%s: ssm_dt_rank      = %u\n",     __func__, hparams.ssm_dt_rank);
-        LLAMA_LOG_INFO("%s: ssm_dt_b_c_rms   = %d\n",     __func__, hparams.ssm_dt_b_c_rms);
-    }
+    // {n_embd, 2*d_inner} @ {n_embd, n_seq_tokens, n_seqs} => {2*d_inner, n_seq_tokens, n_seqs}
+    struct ggml_tensor * xz = llm_build_lora_mm(lctx, ctx, model.layers[il].ssm_in, cur);
+    // split the above in two
+    // => {d_inner, n_seq_tokens, n_seqs}
+    struct ggml_tensor * x = ggml_view_3d(ctx, xz, d_inner, xz->ne[1], xz->ne[2], xz->nb[1], xz->nb[2], 0);
+    struct ggml_tensor * z = ggml_view_3d(ctx, xz, d_inner, xz->ne[1], xz->ne[2], xz->nb[1], xz->nb[2], d_inner*ggml_element_size(xz));
 
-    LLAMA_LOG_INFO("%s: model type       = %s\n",     __func__, llama_model_type_name(model.type));
-    LLAMA_LOG_INFO("%s: model ftype      = %s\n",     __func__, llama_model_ftype_name(model.ftype).c_str());
-    if (ml.n_elements >= 1e12) {
-        LLAMA_LOG_INFO("%s: model params     = %.2f T\n", __func__, ml.n_elements*1e-12);
-    } else if (ml.n_elements >= 1e9) {
-        LLAMA_LOG_INFO("%s: model params     = %.2f B\n", __func__, ml.n_elements*1e-9);
-    } else if (ml.n_elements >= 1e6) {
-        LLAMA_LOG_INFO("%s: model params     = %.2f M\n", __func__, ml.n_elements*1e-6);
-    } else {
-        LLAMA_LOG_INFO("%s: model params     = %.2f K\n", __func__, ml.n_elements*1e-3);
-    }
-    if (ml.n_bytes < GiB) {
-        LLAMA_LOG_INFO("%s: model size       = %.2f MiB (%.2f BPW) \n", __func__, ml.n_bytes/1024.0/1024.0,        ml.n_bytes*8.0/ml.n_elements);
-    } else {
-        LLAMA_LOG_INFO("%s: model size       = %.2f GiB (%.2f BPW) \n", __func__, ml.n_bytes/1024.0/1024.0/1024.0, ml.n_bytes*8.0/ml.n_elements);
-    }
+    // conv
+    {
+        // => {d_conv - 1 + n_seq_tokens, d_inner, n_seqs}
+        struct ggml_tensor * conv_x = ggml_concat(ctx, conv, ggml_transpose(ctx, x), 0);
 
-    // general kv
-    LLAMA_LOG_INFO("%s: general.name     = %s\n",    __func__, model.name.c_str());
-
-    // special tokens
-    if (vocab.special_bos_id  != -1)    { LLAMA_LOG_INFO( "%s: BOS token        = %d '%s'\n", __func__, vocab.special_bos_id,     vocab.id_to_token[vocab.special_bos_id].text.c_str() );  }
-    if (vocab.special_eos_id  != -1)    { LLAMA_LOG_INFO( "%s: EOS token        = %d '%s'\n", __func__, vocab.special_eos_id,     vocab.id_to_token[vocab.special_eos_id].text.c_str() );  }
-    if (vocab.special_eot_id  != -1)    { LLAMA_LOG_INFO( "%s: EOT token        = %d '%s'\n", __func__, vocab.special_eot_id,     vocab.id_to_token[vocab.special_eot_id].text.c_str() );  }
-    if (vocab.special_eom_id  != -1)    { LLAMA_LOG_INFO( "%s: EOM token        = %d '%s'\n", __func__, vocab.special_eom_id,     vocab.id_to_token[vocab.special_eom_id].text.c_str() );  }
-    if (vocab.special_unk_id  != -1)    { LLAMA_LOG_INFO( "%s: UNK token        = %d '%s'\n", __func__, vocab.special_unk_id,     vocab.id_to_token[vocab.special_unk_id].text.c_str() );  }
-    if (vocab.special_sep_id  != -1)    { LLAMA_LOG_INFO( "%s: SEP token        = %d '%s'\n", __func__, vocab.special_sep_id,     vocab.id_to_token[vocab.special_sep_id].text.c_str() );  }
-    if (vocab.special_pad_id  != -1)    { LLAMA_LOG_INFO( "%s: PAD token        = %d '%s'\n", __func__, vocab.special_pad_id,     vocab.id_to_token[vocab.special_pad_id].text.c_str() );  }
-    if (vocab.special_cls_id  != -1)    { LLAMA_LOG_INFO( "%s: CLS token        = %d '%s'\n", __func__, vocab.special_cls_id,     vocab.id_to_token[vocab.special_cls_id].text.c_str() );  }
-    if (vocab.special_mask_id != -1)    { LLAMA_LOG_INFO( "%s: MASK token       = %d '%s'\n", __func__, vocab.special_mask_id,    vocab.id_to_token[vocab.special_mask_id].text.c_str() ); }
-
-    if (vocab.linefeed_id != -1)        { LLAMA_LOG_INFO( "%s: LF token         = %d '%s'\n", __func__, vocab.linefeed_id,        vocab.id_to_token[vocab.linefeed_id].text.c_str() ); }
-
-    if (vocab.special_fim_pre_id != -1) { LLAMA_LOG_INFO( "%s: FIM PRE token    = %d '%s'\n", __func__, vocab.special_fim_pre_id, vocab.id_to_token[vocab.special_fim_pre_id].text.c_str() ); }
-    if (vocab.special_fim_suf_id != -1) { LLAMA_LOG_INFO( "%s: FIM SUF token    = %d '%s'\n", __func__, vocab.special_fim_suf_id, vocab.id_to_token[vocab.special_fim_suf_id].text.c_str() ); }
-    if (vocab.special_fim_mid_id != -1) { LLAMA_LOG_INFO( "%s: FIM MID token    = %d '%s'\n", __func__, vocab.special_fim_mid_id, vocab.id_to_token[vocab.special_fim_mid_id].text.c_str() ); }
-    if (vocab.special_fim_pad_id != -1) { LLAMA_LOG_INFO( "%s: FIM PAD token    = %d '%s'\n", __func__, vocab.special_fim_pad_id, vocab.id_to_token[vocab.special_fim_pad_id].text.c_str() ); }
-    if (vocab.special_fim_rep_id != -1) { LLAMA_LOG_INFO( "%s: FIM REP token    = %d '%s'\n", __func__, vocab.special_fim_rep_id, vocab.id_to_token[vocab.special_fim_rep_id].text.c_str() ); }
-    if (vocab.special_fim_sep_id != -1) { LLAMA_LOG_INFO( "%s: FIM SEP token    = %d '%s'\n", __func__, vocab.special_fim_sep_id, vocab.id_to_token[vocab.special_fim_sep_id].text.c_str() ); }
-
-    for (const auto & id : vocab.special_eog_ids) {
-        LLAMA_LOG_INFO( "%s: EOG token        = %d '%s'\n", __func__, id, vocab.id_to_token[id].text.c_str() );
-    }
+        // copy last (d_conv - 1) columns back into the state cache
+        struct ggml_tensor * last_conv = ggml_view_3d(ctx, conv_x, d_conv - 1, d_inner, n_seqs, conv_x->nb[1], conv_x->nb[2], n_seq_tokens*(conv_x->nb[0]));
 
-    LLAMA_LOG_INFO("%s: max token length = %d\n", __func__, vocab.max_token_len);
+        ggml_build_forward_expand(graph,
+            ggml_cpy(ctx, last_conv,
+                ggml_view_1d(ctx, conv_states_all,
+                    (d_conv - 1)*(d_inner)*(n_seqs),
+                    kv_head*(d_conv - 1)*(d_inner)*ggml_element_size(conv_states_all))));
 
-    if (model.arch == LLM_ARCH_DEEPSEEK) {
-        LLAMA_LOG_INFO("%s: n_layer_dense_lead   = %d\n",     __func__, hparams.n_layer_dense_lead);
-        LLAMA_LOG_INFO("%s: n_ff_exp             = %d\n",     __func__, hparams.n_ff_exp);
-        LLAMA_LOG_INFO("%s: n_expert_shared      = %d\n",     __func__, hparams.n_expert_shared);
-        LLAMA_LOG_INFO("%s: expert_weights_scale = %.1f\n",   __func__, hparams.expert_weights_scale);
-    }
+        // 1D convolution
+        // The equivalent is to make a self-overlapping view of conv_x
+        // over d_conv columns at each stride in the 3rd dimension,
+        // then element-wise multiply that with the conv1d weight,
+        // then sum the elements of each row,
+        // (the last two steps are a dot product over rows (also doable with mul_mat))
+        // then permute away the ne[0] dimension,
+        // and then you're left with the resulting x tensor.
+        // For simultaneous sequences, all sequences need to have the same length.
+        x = ggml_ssm_conv(ctx, conv_x, model.layers[il].ssm_conv1d);
 
-    if (model.arch == LLM_ARCH_DEEPSEEK2) {
-        LLAMA_LOG_INFO("%s: n_layer_dense_lead   = %d\n",     __func__, hparams.n_layer_dense_lead);
-        LLAMA_LOG_INFO("%s: n_lora_q             = %d\n",     __func__, hparams.n_lora_q);
-        LLAMA_LOG_INFO("%s: n_lora_kv            = %d\n",     __func__, hparams.n_lora_kv);
-        LLAMA_LOG_INFO("%s: n_ff_exp             = %d\n",     __func__, hparams.n_ff_exp);
-        LLAMA_LOG_INFO("%s: n_expert_shared      = %d\n",     __func__, hparams.n_expert_shared);
-        LLAMA_LOG_INFO("%s: expert_weights_scale = %.1f\n",   __func__, hparams.expert_weights_scale);
-        LLAMA_LOG_INFO("%s: rope_yarn_log_mul    = %.4f\n",   __func__, hparams.rope_yarn_log_mul);
-    }
+        // bias
+        x = ggml_add(ctx, x, model.layers[il].ssm_conv1d_b);
 
-    if (model.arch == LLM_ARCH_QWEN2MOE) {
-        LLAMA_LOG_INFO("%s: n_ff_exp         = %d\n",     __func__, hparams.n_ff_exp);
-        LLAMA_LOG_INFO("%s: n_ff_shexp       = %d\n",     __func__, hparams.n_ff_shexp);
+        x = ggml_silu(ctx, x);
     }
 
-    if (model.arch == LLM_ARCH_MINICPM || model.arch == LLM_ARCH_GRANITE || model.arch == LLM_ARCH_GRANITE_MOE) {
-        LLAMA_LOG_INFO("%s: f_embedding_scale = %f\n", __func__, hparams.f_embedding_scale);
-        LLAMA_LOG_INFO("%s: f_residual_scale  = %f\n", __func__, hparams.f_residual_scale);
-        LLAMA_LOG_INFO("%s: f_attention_scale = %f\n", __func__, hparams.f_attention_scale);
-    }
-}
-
-enum llm_tensor_layer {
-    LLM_TENSOR_LAYER_INPUT,
-    LLM_TENSOR_LAYER_REPEATING,
-    LLM_TENSOR_LAYER_OUTPUT,
-};
+    // ssm
+    {
+        // {d_inner, dt_rank + 2*d_state} @ {d_inner, n_seq_tokens, n_seqs} => {dt_rank + 2*d_state, n_seq_tokens, n_seqs}
+        struct ggml_tensor * x_db = llm_build_lora_mm(lctx, ctx, model.layers[il].ssm_x, x);
+        // split
+        struct ggml_tensor * dt = ggml_view_3d(ctx, x_db, dt_rank, n_seq_tokens, n_seqs, x_db->nb[1], x_db->nb[2], 0);
+        struct ggml_tensor * B  = ggml_view_3d(ctx, x_db, d_state, n_seq_tokens, n_seqs, x_db->nb[1], x_db->nb[2], ggml_element_size(x_db)*dt_rank);
+        struct ggml_tensor * C  = ggml_view_3d(ctx, x_db, d_state, n_seq_tokens, n_seqs, x_db->nb[1], x_db->nb[2], ggml_element_size(x_db)*(dt_rank+d_state));
 
-struct llm_tensor_info {
-    llm_tensor_layer layer;
-    ggml_op op;
-};
+        // Some Mamba variants (e.g. FalconMamba) apply RMS norm in B, C & Dt layers
+        if (ssm_dt_b_c_rms) {
+            dt = ggml_rms_norm(ctx, dt, norm_rms_eps);
+            B = ggml_rms_norm(ctx, B, norm_rms_eps);
+            C = ggml_rms_norm(ctx, C, norm_rms_eps);
+        }
 
-static const std::map llm_tensor_info_mapping = {
-    {LLM_TENSOR_TOKEN_EMBD,                 {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
-    {LLM_TENSOR_POS_EMBD,                   {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
-    {LLM_TENSOR_TOKEN_EMBD_NORM,            {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
-    {LLM_TENSOR_TOKEN_TYPES,                {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
-    {LLM_TENSOR_OUTPUT,                     {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_CLS,                        {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_CLS_OUT,                    {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_OUTPUT_NORM,                {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
-    {LLM_TENSOR_DEC_OUTPUT_NORM,            {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
-    {LLM_TENSOR_ENC_OUTPUT_NORM,            {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
-    {LLM_TENSOR_ROPE_FREQS,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ROPE}},
-    {LLM_TENSOR_ROPE_FACTORS_LONG,          {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ROPE}},
-    {LLM_TENSOR_ROPE_FACTORS_SHORT,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ROPE}},
-    {LLM_TENSOR_ATTN_Q,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_K,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_V,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_QKV,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_OUT,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_GATE,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_DOWN,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_UP,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_DOWN_SHEXP,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_GATE_SHEXP,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_UP_SHEXP,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_Q_A,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_Q_B,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_KV_A_MQA,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_KV_B,                  {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_ATTN_Q,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_ATTN_K,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_Q,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_K,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_V,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_QKV,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_OUT,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_GATE,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_DOWN,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_UP,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_DOWN_SHEXP,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_GATE_SHEXP,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_UP_SHEXP,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_Q_A,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_Q_B,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_KV_A_MQA,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ATTN_KV_B,                  {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_ATTN_Q,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_ATTN_K,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_ATTN_V,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_ATTN_OUT,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_CROSS_ATTN_Q,           {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_CROSS_ATTN_K,           {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_CROSS_ATTN_V,           {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_CROSS_ATTN_OUT,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_FFN_GATE,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_FFN_DOWN,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_DEC_FFN_UP,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ENC_ATTN_Q,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ENC_ATTN_K,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ENC_ATTN_V,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ENC_ATTN_OUT,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ENC_FFN_GATE,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ENC_FFN_DOWN,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_ENC_FFN_UP,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_GATE_INP_SHEXP,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_GATE_INP,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_SSM_IN,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_SSM_X,                      {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_SSM_DT,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_SSM_OUT,                    {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_TIME_MIX_W1,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_TIME_MIX_W2,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_TIME_MIX_DECAY_W1,          {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_TIME_MIX_DECAY_W2,          {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_TIME_MIX_KEY,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_TIME_MIX_VALUE,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_TIME_MIX_RECEPTANCE,        {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_TIME_MIX_GATE,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_TIME_MIX_OUTPUT,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_CHANNEL_MIX_KEY,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_CHANNEL_MIX_RECEPTANCE,     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_CHANNEL_MIX_VALUE,          {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
-    {LLM_TENSOR_FFN_ACT,                    {LLM_TENSOR_LAYER_REPEATING, GGML_OP_DIV}},
-    {LLM_TENSOR_SSM_CONV1D,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_SSM_CONV}},
-    {LLM_TENSOR_SSM_A,                      {LLM_TENSOR_LAYER_REPEATING, GGML_OP_SSM_SCAN}},
-    {LLM_TENSOR_SSM_D,                      {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_TIME_MIX_LERP_X,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_TIME_MIX_LN,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_CHANNEL_MIX_LERP_K,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_CHANNEL_MIX_LERP_R,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_TIME_MIX_LERP_W,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
-    {LLM_TENSOR_TIME_MIX_LERP_K,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
-    {LLM_TENSOR_TIME_MIX_LERP_V,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
-    {LLM_TENSOR_TIME_MIX_LERP_R,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
-    {LLM_TENSOR_TIME_MIX_LERP_G,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
-    {LLM_TENSOR_TIME_MIX_DECAY,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
-    {LLM_TENSOR_TIME_MIX_FIRST,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_RWKV_WKV6}},
-    {LLM_TENSOR_ATTN_NORM,                  {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_ATTN_NORM_2,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_ATTN_OUT_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_ATTN_POST_NORM,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_FFN_NORM,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_FFN_POST_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_FFN_NORM_EXPS,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_ATTN_Q_NORM,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_ATTN_K_NORM,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_LAYER_OUT_NORM,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_ATTN_Q_A_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_ATTN_KV_A_NORM,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_ATTN_SUB_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_FFN_SUB_NORM,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_DEC_ATTN_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_DEC_CROSS_ATTN_NORM,        {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_DEC_FFN_NORM,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_ENC_ATTN_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_ENC_FFN_NORM,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
-    {LLM_TENSOR_DEC_ATTN_REL_B,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_GET_ROWS}},
-    {LLM_TENSOR_ENC_ATTN_REL_B,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_GET_ROWS}},
-    {LLM_TENSOR_FFN_DOWN_EXPS,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
-    {LLM_TENSOR_FFN_GATE_EXPS,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
-    {LLM_TENSOR_FFN_UP_EXPS,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
-    // this tensor is loaded for T5, but never used
-    {LLM_TENSOR_DEC_CROSS_ATTN_REL_B,       {LLM_TENSOR_LAYER_REPEATING, GGML_OP_NONE}},
-};
+        // {dt_rank, d_inner} @ {dt_rank, n_seq_tokens, n_seqs} => {d_inner, n_seq_tokens, n_seqs}
+        dt = llm_build_lora_mm(lctx, ctx, model.layers[il].ssm_dt, dt);
+        dt = ggml_add(ctx, dt, model.layers[il].ssm_dt_b);
 
-// checks if the weight tensor can be used with the specified buffer type and device
-static bool weight_buft_supported(const llama_hparams & hparams, ggml_tensor * w, ggml_op op, ggml_backend_buffer_type_t buft, ggml_backend_dev_t dev) {
-    GGML_ASSERT(w != nullptr);
+        // Custom operator to optimize the parallel associative scan
+        // as described in the Annex D of the Mamba paper.
+        // => {d_inner, n_seq_tokens, n_seqs} and {d_state, d_inner, n_seqs}
+        struct ggml_tensor * y_ssm = ggml_ssm_scan(ctx, ssm, x, dt, model.layers[il].ssm_a, B, C);
 
-    if (op == GGML_OP_NONE) {
-        return true;
-    }
+        // store last states
+        ggml_build_forward_expand(graph,
+            ggml_cpy(ctx,
+                ggml_view_1d(ctx, y_ssm, d_state*d_inner*n_seqs, x->nb[3]),
+                ggml_view_1d(ctx, ssm_states_all, d_state*d_inner*n_seqs, kv_head*d_state*d_inner*ggml_element_size(ssm_states_all))));
 
-    ggml_init_params params = {
-        /*.mem_size   =*/ ggml_tensor_overhead()*8,
-        /*.mem_buffer =*/ NULL,
-        /*.no_alloc   =*/ true,
-    };
-    ggml_context_ptr ctx_ptr { ggml_init(params) };
-    if (!ctx_ptr) {
-        throw std::runtime_error(format("failed to create ggml context"));
-    }
-    ggml_context * ctx = ctx_ptr.get();
+        struct ggml_tensor * y = ggml_view_3d(ctx, y_ssm, d_inner, n_seq_tokens, n_seqs, x->nb[1], x->nb[2], 0);
 
-    ggml_tensor * op_tensor = nullptr;
+        // TODO: skip computing output earlier for unused tokens
 
-    switch (op) {
-        case GGML_OP_GET_ROWS:
-            {
-                ggml_tensor * b = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, 512);
-                op_tensor = ggml_get_rows(ctx, w, b);
-            } break;
-        case GGML_OP_MUL_MAT:
-            {
-                ggml_tensor * b = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, w->ne[0], 512, w->ne[2], w->ne[3]);
-                op_tensor = ggml_mul_mat(ctx, w, b);
-            } break;
-        case GGML_OP_MUL_MAT_ID:
-            {
-                int n_expert_used = hparams.n_expert_used;
-                ggml_tensor * b = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, w->ne[0], n_expert_used, 512);
-                ggml_tensor * ids = ggml_new_tensor_2d(ctx, GGML_TYPE_I32, n_expert_used, 512);
-                op_tensor = ggml_mul_mat_id(ctx, w, b, ids);
-            } break;
-        case GGML_OP_ADD:
-            {
-                ggml_tensor * a = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, w->ne[0], w->ne[1], w->ne[2], w->ne[3]);
-                op_tensor = ggml_add(ctx, a, w);
-            } break;
-        case GGML_OP_MUL:
-            {
-                ggml_tensor * a = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, w->ne[0], w->ne[1], w->ne[2], w->ne[3]);
-                op_tensor = ggml_mul(ctx, a, w);
-            } break;
-        case GGML_OP_DIV:
-            {
-                ggml_tensor * a = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, w->ne[0]);
-                op_tensor = ggml_div(ctx, a, w);
-            } break;
-        case GGML_OP_ROPE:
-            {
-                int n_embd_head = hparams.n_embd_head_v;
-                int n_head = hparams.n_head();
-                ggml_tensor * a = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, n_embd_head, n_head, 512);
-                ggml_tensor * b = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, 512);
-                op_tensor = ggml_rope_ext(
-                    ctx, a, b, w,
-                    0, 0, 0, 0, 0,
-                    0, 0, 0, 0
-                );
+        // {d_inner, n_seq_tokens, n_seqs} * {d_inner} => {d_inner, n_seq_tokens, n_seqs}
+        y = ggml_add(ctx, y, ggml_mul(ctx, x, model.layers[il].ssm_d));
+        y = ggml_mul(ctx, y, ggml_silu(ctx, ggml_cont(ctx, z)));
 
-            } break;
-        case GGML_OP_SSM_CONV:
-            {
-                // FIXME
-                ggml_tensor * conv_x = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, 12345, w->ne[1], 6789);
-                op_tensor = ggml_ssm_conv(ctx, conv_x, w);
-            } break;
-        case GGML_OP_SSM_SCAN:
-            {
-                // FIXME
-                const int64_t d_state      = w->ne[0];
-                const int64_t d_inner      = w->ne[1];
-                const int64_t n_seq_tokens = 512;
-                const int64_t n_seqs       = 1;
-                ggml_tensor * s  = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, d_state, d_inner, n_seqs);
-                ggml_tensor * x = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, d_inner, n_seq_tokens, n_seqs);
-                ggml_tensor * dt = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, d_inner, n_seq_tokens, n_seqs);
-                ggml_tensor * B = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, d_state, n_seq_tokens, n_seqs);
-                ggml_tensor * C = ggml_new_tensor_3d(ctx, GGML_TYPE_F32, d_state, n_seq_tokens, n_seqs);
-                op_tensor = ggml_ssm_scan(ctx, s, x, dt, w, B, C);
-            } break;
-        case GGML_OP_RWKV_WKV6:
-            {
-                // FIXME
-                const int64_t S = 123;
-                const int64_t H = 123;
-                const int64_t n_tokens = 123;
-                const int64_t n_seqs = 123;
-                ggml_tensor  * k = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, S, 1, H, n_tokens);
-                ggml_tensor  * v = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, 1, S, H, n_tokens);
-                ggml_tensor  * r = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, 1, S, H, n_tokens);
-                ggml_tensor  * tf = w;
-                ggml_tensor  * td = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, 1, S, H, n_tokens);
-                ggml_tensor  * state = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, S, n_seqs, S, H);
-                op_tensor = ggml_rwkv_wkv6(ctx, k, v, r, tf, td, state);
-            } break;
-        default:
-            GGML_ABORT("%s: missing test for op %s for tensor %s", __func__, ggml_op_name(op), w->name);
+        // {d_inner, n_embd} @ {d_inner, n_seq_tokens, n_seqs} => {n_embd, n_seq_tokens, n_seqs}
+        cur = llm_build_lora_mm(lctx, ctx, model.layers[il].ssm_out, y);
     }
 
-    // create a temporary dummy buffer for the weight so that supports_op can check the buffer type
-    GGML_ASSERT(w->buffer == nullptr);
-    w->buffer = ggml_backend_buft_alloc_buffer(buft, 0);
-    bool op_supported = ggml_backend_dev_supports_op(dev, op_tensor);
-    ggml_backend_buffer_free(w->buffer);
-    w->buffer = nullptr;
+    // {n_embd, n_seq_tokens, n_seqs} => {n_embd, n_tokens}
+    cur = ggml_reshape_2d(ctx, cur, cur->ne[0], n_seq_tokens * n_seqs);
+    cb(cur, "mamba_out", il);
 
-    return op_supported;
+    return cur;
 }
 
-// find the first buffer type in the list that can use the tensor
-static ggml_backend_buffer_type_t select_weight_buft(const llama_model & model, ggml_tensor * tensor, ggml_op op, const llama_model::buft_list_t & buft_list) {
-    GGML_ASSERT(!buft_list.empty());
-    for (const auto & cur : buft_list) {
-        ggml_backend_dev_t cur_dev = cur.first;
-        ggml_backend_buffer_type_t cur_buft = cur.second;
-        if (weight_buft_supported(model.hparams, tensor, op, cur_buft, cur_dev)) {
-            return cur_buft;
-        }
-    }
-    return nullptr;
-}
+static struct ggml_tensor * llm_build_rwkv6_time_mix(
+        struct llama_context & lctx,
+        struct ggml_context * ctx,
+        const struct llama_layer * layer,
+        struct ggml_tensor * cur,
+        struct ggml_tensor * x_prev,
+        struct ggml_tensor ** wkv_state) {
+    size_t n_embd       = cur->ne[0];
+    size_t n_seq_tokens = cur->ne[1];
+    size_t n_seqs       = cur->ne[2];
 
-// CPU: ACCEL -> CPU extra -> GPU host -> CPU
-static llama_model::buft_list_t make_cpu_buft_list(llama_model & model) {
-    llama_model::buft_list_t buft_list;
+    size_t head_size  = layer->time_mix_first->ne[0];
+    size_t head_count = layer->time_mix_first->ne[1];
 
-    // add ACCEL buffer types
-    for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
-        ggml_backend_dev_t dev = ggml_backend_dev_get(i);
-        if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_ACCEL) {
-            auto * buft = ggml_backend_dev_buffer_type(dev);
-            // skip
-            if (buft != ggml_backend_cpu_buffer_type()) {
-                buft_list.emplace_back(dev, buft);
-            }
-        }
-    }
+    size_t n_tokens = n_seqs * n_seq_tokens;
 
-    // add extra buffer types
-    auto * cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
-    auto * cpu_reg = ggml_backend_dev_backend_reg(cpu_dev);
-    auto ggml_backend_dev_get_extra_bufts_fn = (ggml_backend_dev_get_extra_bufts_t)
-        ggml_backend_reg_get_proc_address(cpu_reg, "ggml_backend_dev_get_extra_bufts");
-    if (ggml_backend_dev_get_extra_bufts_fn) {
-        ggml_backend_buffer_type_t * extra_bufts = ggml_backend_dev_get_extra_bufts_fn(cpu_dev);
-        while (extra_bufts && *extra_bufts) {
-            buft_list.emplace_back(cpu_dev, *extra_bufts);
-            ++extra_bufts;
-        }
-    }
+    struct ggml_tensor * sx = ggml_sub(ctx, x_prev, cur);
 
-    // add a host buffer type
-    // storing the tensors in a host buffer is useful when the processing of large batches
-    // is offloaded to a GPU device, since it reduces the time spent on data transfers
-    // generally, this will be done using the first device in the list
-    // a better approach would be to handle this on a weight-by-weight basis using the offload_op
-    // function of the device to determine if it would benefit from being stored in a host buffer
-    for (auto * dev : model.devices) {
-        ggml_backend_buffer_type_t buft = ggml_backend_dev_host_buffer_type(dev);
-        if (buft) {
-            buft_list.emplace_back(dev, buft);
-            break;
-        }
-    }
+    sx  = ggml_reshape_2d(ctx, sx,  n_embd, n_tokens);
+    cur = ggml_reshape_2d(ctx, cur, n_embd, n_tokens);
 
-    // add the CPU buffer type
-    for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
-        ggml_backend_dev_t dev = ggml_backend_dev_get(i);
-        if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_CPU) {
-            buft_list.emplace_back(dev, ggml_backend_dev_buffer_type(dev));
-        }
-    }
+    struct ggml_tensor * xxx = ggml_add(ctx, ggml_mul(ctx, sx, layer->time_mix_lerp_x), cur);
 
-    return buft_list;
-}
+    xxx = ggml_reshape_4d(
+        ctx,
+        ggml_tanh(
+            ctx,
+            ggml_mul_mat(ctx, layer->time_mix_w1, xxx)
+        ),
+        layer->time_mix_w1->ne[1] / 5, 1, 5, n_tokens
+    );
 
-// GPU: split if LLAMA_SPLIT_MODE_ROW -> GPU
-static llama_model::buft_list_t make_gpu_buft_list(ggml_backend_dev_t dev, enum llama_split_mode split_mode, const float * tensor_split) {
-    llama_model::buft_list_t buft_list;
+    xxx = ggml_cont(ctx, ggml_permute(ctx, xxx, 0, 1, 3, 2));
 
-    // add the device split buffer type if requested and available
-    if (split_mode == LLAMA_SPLIT_MODE_ROW) {
-        ggml_backend_reg_t reg = ggml_backend_dev_backend_reg(dev);
-        auto ggml_backend_split_buffer_type_fn = (ggml_backend_split_buffer_type_t)
-            ggml_backend_reg_get_proc_address(reg, "ggml_backend_split_buffer_type");
-        if (ggml_backend_split_buffer_type_fn) {
-            size_t dev_index = [&]() {
-                auto * reg = ggml_backend_dev_backend_reg(dev);
-                for (size_t i = 0; i < ggml_backend_reg_dev_count(reg); ++i) {
-                    if (ggml_backend_reg_dev_get(reg, i) == dev) {
-                        return i;
-                    }
-                }
-                throw std::runtime_error(format("device %s not found in its backend reg", ggml_backend_dev_name(dev)));
-            }();
-            auto * buft = ggml_backend_split_buffer_type_fn(dev_index, tensor_split);
-            if (buft != nullptr) {
-                buft_list.emplace_back(dev, buft);
-            }
-        }
-    }
-
-    // add the device default buffer type
-    buft_list.emplace_back(dev, ggml_backend_dev_buffer_type(dev));
-
-    return buft_list;
-}
+    xxx = ggml_mul_mat(
+        ctx,
+        ggml_reshape_4d(
+            ctx,
+            layer->time_mix_w2,
+            layer->time_mix_w2->ne[0], layer->time_mix_w2->ne[1], 1, 5
+        ),
+        xxx
+    );
 
-// Returns false if cancelled by progress_callback
-static bool llm_load_tensors(
-        llama_model_loader & ml,
-        llama_model & model,
-        int n_gpu_layers,
-        enum llama_split_mode split_mode,
-        int main_gpu,
-        const float * tensor_split,
-        bool use_mlock,
-        llama_progress_callback progress_callback,
-        void * progress_callback_user_data) {
-    auto & hparams = model.hparams;
+    struct ggml_tensor *mw = ggml_view_2d(ctx, xxx, n_embd, n_tokens, xxx->nb[1], 0);
+    struct ggml_tensor *mk = ggml_view_2d(ctx, xxx, n_embd, n_tokens, xxx->nb[1], n_embd * n_tokens * sizeof(float));
+    struct ggml_tensor *mv = ggml_view_2d(ctx, xxx, n_embd, n_tokens, xxx->nb[1], n_embd * n_tokens * 2 * sizeof(float));
+    struct ggml_tensor *mr = ggml_view_2d(ctx, xxx, n_embd, n_tokens, xxx->nb[1], n_embd * n_tokens * 3 * sizeof(float));
+    struct ggml_tensor *mg = ggml_view_2d(ctx, xxx, n_embd, n_tokens, xxx->nb[1], n_embd * n_tokens * 4 * sizeof(float));
 
-    model.split_mode   = split_mode;
-    model.main_gpu     = main_gpu;
-    model.n_gpu_layers = n_gpu_layers;
+    struct ggml_tensor * xw = ggml_add(
+        ctx,
+        ggml_mul(
+            ctx,
+            ggml_add(ctx, mw, layer->time_mix_lerp_w),
+            sx
+        ),
+        cur
+    );
 
-    const int n_layer     = hparams.n_layer;
-    bool use_mmap_buffer = true;
+    struct ggml_tensor * xk = ggml_add(
+        ctx,
+        ggml_mul(
+            ctx,
+            ggml_add(ctx, mk, layer->time_mix_lerp_k),
+            sx
+        ),
+        cur
+    );
 
-    // build a list of buffer types for the CPU and GPU devices
-    model.cpu_buft_list = make_cpu_buft_list(model);
-    for (auto * dev : model.devices) {
-        llama_model::buft_list_t buft_list = make_gpu_buft_list(dev, split_mode, tensor_split);
-        // add CPU buffer types as a fallback
-        buft_list.insert(buft_list.end(), model.cpu_buft_list.begin(), model.cpu_buft_list.end());
-        model.gpu_buft_list.emplace(dev, std::move(buft_list));
-    }
+    struct ggml_tensor * xv = ggml_add(
+        ctx,
+        ggml_mul(
+            ctx,
+            ggml_add(ctx, mv, layer->time_mix_lerp_v),
+            sx
+        ),
+        cur
+    );
 
-    // calculate the split points
-    int device_count = llama_get_device_count(model);
-    bool all_zero = tensor_split == nullptr || std::all_of(tensor_split, tensor_split + device_count, [](float x) { return x == 0.0f; });
-    std::vector splits(device_count);
-    if (all_zero) {
-        // default split, by free memory
-        for (int i = 0; i < device_count; ++i) {
-            ggml_backend_dev_t dev = model.devices[i];
-            size_t total;
-            size_t free;
-            ggml_backend_dev_memory(dev, &free, &total);
-            splits[i] = free;
-        }
-    } else {
-        std::copy(tensor_split, tensor_split + device_count, splits.begin());
-    }
+    struct ggml_tensor * xr = ggml_add(
+        ctx,
+        ggml_mul(
+            ctx,
+            ggml_add(ctx, mr, layer->time_mix_lerp_r),
+            sx
+        ),
+        cur
+    );
 
-    // sum and normalize the splits to get the split points
-    float split_sum = 0.0f;
-    for (int i = 0; i < device_count; ++i) {
-        split_sum += splits[i];
-        splits[i] = split_sum;
-    }
-    for (int i = 0; i < device_count; ++i) {
-        splits[i] /= split_sum;
-    }
+    struct ggml_tensor * xg = ggml_add(
+        ctx,
+        ggml_mul(
+            ctx,
+            ggml_add(ctx, mg, layer->time_mix_lerp_g),
+            sx
+        ),
+        cur
+    );
 
-    ggml_backend_dev_t cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
-    const int i_gpu_start = std::max((int) hparams.n_layer - n_gpu_layers, (int) 0);
-    const int act_gpu_layers = model.devices.empty() ? 0 : std::min(n_gpu_layers, (int)n_layer + 1);
-    auto get_layer_buft_list = [&](int il) -> llama_model::layer_dev {
-        if (il < i_gpu_start || (il - i_gpu_start) >= act_gpu_layers) {
-            return {cpu_dev, &model.cpu_buft_list};
-        }
-        int layer_gpu = std::upper_bound(splits.begin(), splits.begin() + device_count, float(il - i_gpu_start)/act_gpu_layers) - splits.begin();
-        auto * dev = model.devices.at(layer_gpu);
-        return {dev, &model.gpu_buft_list.at(dev)};
-    };
+    struct ggml_tensor * r = ggml_reshape_4d(ctx, llm_build_lora_mm(lctx, ctx, layer->time_mix_receptance, xr), head_size, 1,         head_count, n_tokens);
+    struct ggml_tensor * k = ggml_reshape_4d(ctx, llm_build_lora_mm(lctx, ctx, layer->time_mix_key,        xk), 1,         head_size, head_count, n_tokens);
+    struct ggml_tensor * v = ggml_reshape_4d(ctx, llm_build_lora_mm(lctx, ctx, layer->time_mix_value,      xv), head_size, 1,         head_count, n_tokens);
+    struct ggml_tensor * g = ggml_silu(
+        ctx,
+        llm_build_lora_mm(lctx, ctx, layer->time_mix_gate, xg)
+    );
 
-    // assign the input layer
-    // there is very little benefit to offloading the input layer, so always keep it on the CPU
-    model.dev_input = { cpu_dev, &model.cpu_buft_list };
+    struct ggml_tensor * w = ggml_mul_mat(
+        ctx,
+        layer->time_mix_decay_w2,
+        ggml_tanh(
+            ctx,
+            ggml_mul_mat(ctx, layer->time_mix_decay_w1, xw)
+        )
+    );
 
-    // assign the repeating layers to the devices according to the splits
-    model.dev_layer.resize(n_layer);
-    for (int il = 0; il < n_layer; ++il) {
-        model.dev_layer[il] = get_layer_buft_list(il);
-    }
-    // assign the output layer
-    model.dev_output = get_layer_buft_list(n_layer);
+    w = ggml_add(ctx, w, ggml_reshape_1d(ctx, layer->time_mix_decay, n_embd));
+    w = ggml_exp(ctx, ggml_neg(ctx, ggml_exp(ctx, w)));
+    w = ggml_reshape_4d(ctx, w, 1, head_size, head_count, n_tokens);
 
-    // one ggml context per buffer type
-    int max_n_tensors = ml.n_tensors;
-    max_n_tensors += 1;         // duplicated output tensor
-    max_n_tensors += n_layer*2; // duplicated rope freq tensors
-    const size_t ctx_size = ggml_tensor_overhead()*max_n_tensors;
+    k = ggml_transpose(ctx, k);
+    v = ggml_transpose(ctx, v);
+    r = ggml_transpose(ctx, r);
 
-    std::map ctx_map;
-    auto ctx_for_buft = [&](ggml_backend_buffer_type_t buft) -> ggml_context * {
-        auto it = ctx_map.find(buft);
-        if (it == ctx_map.end()) {
-            ggml_init_params params = {
-                /*.mem_size   =*/ ctx_size,
-                /*.mem_buffer =*/ NULL,
-                /*.no_alloc   =*/ true,
-            };
-            ggml_context * ctx = ggml_init(params);
-            if (!ctx) {
-                throw std::runtime_error(format("failed to create ggml context"));
-            }
-            ctx_map[buft] = ctx;
-            model.ctxs.emplace_back(ctx);
-            return ctx;
-        }
-        return it->second;
-    };
+    struct ggml_tensor * wkv_output = ggml_rwkv_wkv6(ctx, k, v, r, layer->time_mix_first, w, *wkv_state);
+    cur = ggml_view_1d(ctx, wkv_output, n_embd * n_tokens, 0);
+    *wkv_state = ggml_view_1d(ctx, wkv_output, n_embd * head_size * n_seqs, n_embd * n_tokens * sizeof(float));
 
-    // create tensors for the weights
-    {
-        // note: cast to int64_t since we will use these for the tensor dimensions
-        const int64_t n_head        = hparams.n_head();
-        const int64_t n_head_kv     = hparams.n_head_kv();
-        const int64_t n_embd        = hparams.n_embd;
-        const int64_t n_embd_k_gqa  = hparams.n_embd_k_gqa();
-        const int64_t n_embd_v_gqa  = hparams.n_embd_v_gqa();
-        const int64_t n_embd_head_k = hparams.n_embd_head_k;
-        const int64_t n_embd_head_v = hparams.n_embd_head_v;
-        const int64_t n_ff          = hparams.n_ff();
-        const int64_t n_embd_gqa    = n_embd_v_gqa;
-        const int64_t n_vocab       = hparams.n_vocab;
-        const int64_t n_vocab_type  = hparams.n_vocab_type;
-        const int64_t n_rot         = hparams.n_rot;
-        const int64_t n_expert      = hparams.n_expert;
-        const int64_t n_expert_used = hparams.n_expert_used;
-        const int64_t n_ctx_train   = hparams.n_ctx_train;
+    // group norm with head_count groups
+    cur = ggml_reshape_3d(ctx, cur, n_embd / head_count, head_count, n_tokens);
+    cur = ggml_norm(ctx, cur, 64e-5f);
 
-        if (n_expert > 0 && hparams.n_expert_used == 0) {
-            throw std::runtime_error("model has expert layers but no expert layers are used");
-        }
+    // Convert back to regular vectors.
+    cur = ggml_reshape_2d(ctx, cur, n_embd, n_tokens);
+    cur = ggml_add(ctx, ggml_mul(ctx, cur, layer->time_mix_ln), layer->time_mix_ln_b);
 
-        int n_moved_tensors = 0;
-        ggml_tensor * first_moved_tensor = nullptr;
-        ggml_backend_buffer_type_t first_moved_from_buft = nullptr;
-        ggml_backend_buffer_type_t first_moved_to_buft = nullptr;
+    cur = ggml_mul(ctx, cur, g);
+    cur = llm_build_lora_mm(lctx, ctx, layer->time_mix_output, cur);
 
-        auto create_tensor = [&](const LLM_TN_IMPL & tn, const std::initializer_list & ne, int flags) -> ggml_tensor * {
-            ggml_tensor * t_meta = ml.get_tensor_meta(tn.str().c_str());
+    return ggml_reshape_3d(ctx, cur, n_embd, n_seq_tokens, n_seqs);
+}
 
-            if (!t_meta) {
-                if (flags & llama_model_loader::TENSOR_NOT_REQUIRED) {
-                    return nullptr;
-                }
-                throw std::runtime_error(format("missing tensor '%s'", tn.str().c_str()));
-            }
+static struct ggml_tensor * llm_build_rwkv6_channel_mix(
+        struct llama_context & lctx,
+        struct ggml_context * ctx,
+        const struct llama_layer * layer,
+        struct ggml_tensor * cur,
+        struct ggml_tensor * x_prev) {
+    struct ggml_tensor * sx = ggml_sub(ctx, x_prev, cur);
+    struct ggml_tensor * xk = ggml_add(ctx, ggml_mul(ctx, sx, layer->channel_mix_lerp_k), cur);
+    struct ggml_tensor * xr = ggml_add(ctx, ggml_mul(ctx, sx, layer->channel_mix_lerp_r), cur);
 
-            // some models use the token embedding tensor as the output, but since these are used in different layers and with different ops
-            // the tensor is duplicated
-            // to handle this, we check if the tensor is duplicated, and if so, we assume that it is being loaded as the output tensor
-            llm_tensor tn_tensor = tn.tensor;
-            if (tn.tensor == LLM_TENSOR_TOKEN_EMBD && flags & llama_model_loader::TENSOR_DUPLICATED) {
-                tn_tensor = LLM_TENSOR_OUTPUT;
-            }
+    struct ggml_tensor * r = ggml_sigmoid(ctx, llm_build_lora_mm(lctx, ctx, layer->channel_mix_receptance, xr));
+    struct ggml_tensor * k = ggml_sqr(
+        ctx,
+        ggml_relu(
+            ctx,
+            llm_build_lora_mm(lctx, ctx, layer->channel_mix_key, xk)
+        )
+    );
 
-            auto it = llm_tensor_info_mapping.find(tn_tensor);
-            if (it == llm_tensor_info_mapping.end()) {
-                throw std::runtime_error(format("missing tensor info mapping for %s", tn.str().c_str()));
-            }
-            const auto & info = it->second;
+    return ggml_mul(ctx, r, llm_build_lora_mm(lctx, ctx, layer->channel_mix_value, k));
+}
 
-            // tensors with "bias" suffix are always used with GGML_OP_ADD
-            ggml_op op;
-            bool bias = tn.suffix != nullptr && strcmp(tn.suffix, "bias") == 0;
-            if (bias) {
-                op = GGML_OP_ADD;
-            } else {
-                op = info.op;
-            }
-
-            // sanity checks
-            if (info.layer == LLM_TENSOR_LAYER_INPUT || info.layer == LLM_TENSOR_LAYER_OUTPUT) {
-                if (tn.bid != -1) {
-                    GGML_ABORT("input/output layer tensor %s used with a layer number", tn.str().c_str());
-                }
-            } else {
-                if (tn.bid == -1) {
-                    GGML_ABORT("repeating layer tensor %s used without a layer number", tn.str().c_str());
-                }
-            }
+struct llm_build_context {
+    const llama_model    & model;
+          llama_context  & lctx;
+    const llama_hparams  & hparams;
+    const llama_cparams  & cparams;
+    const llama_ubatch   & ubatch;
+    const llama_kv_cache & kv_self;
 
-            // select the buffer type for this tensor
-            llama_model::buft_list_t * buft_list;
-            switch (info.layer) {
-                case LLM_TENSOR_LAYER_INPUT:
-                    buft_list = model.dev_input.buft_list;
-                    break;
-                case LLM_TENSOR_LAYER_OUTPUT:
-                    buft_list = model.dev_output.buft_list;
-                    break;
-                case LLM_TENSOR_LAYER_REPEATING:
-                    buft_list = model.dev_layer.at(tn.bid).buft_list;
-                    break;
-                default:
-                    GGML_ABORT("invalid layer %d for tensor %s", info.layer, tn.str().c_str());
-            }
+    const int64_t n_embd;
+    const int64_t n_layer;
+    const int64_t n_rot;
+    const int64_t n_ctx;       // user-specified context size (can be different from n_ctx_train)
+    const int64_t n_head;
+    const int64_t n_head_kv;
+    const int64_t n_embd_head_k;
+    const int64_t n_embd_k_gqa;
+    const int64_t n_embd_head_v;
+    const int64_t n_embd_v_gqa;
+    const int64_t n_expert;
+    const int64_t n_expert_used;
 
-            ggml_backend_buffer_type_t buft = select_weight_buft(model, t_meta, op, *buft_list);
-            if (!buft) {
-                throw std::runtime_error(format("failed to find a compatible buffer type for tensor %s", tn.str().c_str()));
-            }
+    const float freq_base;
+    const float freq_scale;
+    const float ext_factor;
+    const float attn_factor;
+    const float beta_fast;
+    const float beta_slow;
+    const float norm_eps;
+    const float norm_rms_eps;
 
-            // avoid using a host buffer when using mmap
-            auto * buft_dev = ggml_backend_buft_get_device(buft);
-            if (ml.use_mmap && buft_dev && buft == ggml_backend_dev_host_buffer_type(buft_dev)) {
-                auto * cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
-                buft = ggml_backend_dev_buffer_type(cpu_dev);
-            }
+    const int32_t n_tokens;
+    const int32_t n_kv;     // size of KV cache to consider (n_kv <= kv_self.size)
+    const int32_t n_outputs;
+    const int32_t n_outputs_enc;
+    const int32_t kv_head;  // index of where we store new KV data in the cache
+    const int32_t n_ctx_orig;
 
-            if (buft != buft_list->front().second) {
-                n_moved_tensors++;
-                if (!first_moved_tensor) {
-                    first_moved_tensor = t_meta;
-                    first_moved_from_buft = buft_list->front().second;
-                    first_moved_to_buft   = buft;
-                }
-            }
+    const bool flash_attn;
 
-            ggml_context * ctx = ctx_for_buft(buft);
+    const enum llama_pooling_type pooling_type;
+    const enum llama_rope_type    rope_type;
 
-            // if duplicated, check if the original tensor was allocated in the same buffer type context and avoid creating a new one
-            if (flags & llama_model_loader::TENSOR_DUPLICATED) {
-                ggml_tensor * t = ggml_get_tensor(ctx, tn.str().c_str());
-                if (t) {
-                    return t;
-                }
-            }
-            return ml.create_tensor(ctx, tn, ne, flags);
-        };
+    const llm_build_cb & cb;
 
-        model.layers.resize(n_layer);
+    std::vector & buf_compute_meta;
 
-        // TODO: move to a separate function
-        const auto tn = LLM_TN(model.arch);
-        switch (model.arch) {
-            case LLM_ARCH_LLAMA:
-            case LLM_ARCH_REFACT:
-            case LLM_ARCH_MINICPM:
-            case LLM_ARCH_GRANITE:
-            case LLM_ARCH_GRANITE_MOE:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+    struct ggml_context * ctx0 = nullptr;
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
+    // TODO: consider making the entire interface noexcept
+    llm_build_context(
+        llama_context  & lctx,
+    const llama_ubatch & ubatch,
+    const llm_build_cb & cb,
+                  bool   worst_case) :
+        model            (lctx.model),
+        lctx             (lctx),
+        hparams          (model.hparams),
+        cparams          (lctx.cparams),
+        ubatch           (ubatch),
+        kv_self          (lctx.kv_self),
+        n_embd           (hparams.n_embd),
+        n_layer          (hparams.n_layer),
+        n_rot            (hparams.n_rot),
+        n_ctx            (cparams.n_ctx),
+        n_head           (hparams.n_head()),
+        n_head_kv        (hparams.n_head_kv()),
+        n_embd_head_k    (hparams.n_embd_head_k),
+        n_embd_k_gqa     (hparams.n_embd_k_gqa()),
+        n_embd_head_v    (hparams.n_embd_head_v),
+        n_embd_v_gqa     (hparams.n_embd_v_gqa()),
+        n_expert         (hparams.n_expert),
+        n_expert_used    (hparams.n_expert_used),
+        freq_base        (cparams.rope_freq_base),
+        freq_scale       (cparams.rope_freq_scale),
+        ext_factor       (cparams.yarn_ext_factor),
+        attn_factor      (cparams.yarn_attn_factor),
+        beta_fast        (cparams.yarn_beta_fast),
+        beta_slow        (cparams.yarn_beta_slow),
+        norm_eps         (hparams.f_norm_eps),
+        norm_rms_eps     (hparams.f_norm_rms_eps),
+        n_tokens         (ubatch.n_tokens),
+        n_kv             (worst_case ? kv_self.size : kv_self.n),
+        n_outputs        (worst_case ? n_tokens : lctx.n_outputs),
+        n_outputs_enc    (worst_case ? n_tokens : lctx.embd_enc.size() / hparams.n_embd),
+        kv_head          (worst_case ? (kv_self.recurrent ? 0 : kv_self.size - n_tokens) : kv_self.head),
+        n_ctx_orig       (cparams.n_ctx_orig_yarn),
+        flash_attn       (cparams.flash_attn),
+        pooling_type     (cparams.pooling_type),
+        rope_type        (hparams.rope_type),
+        cb               (cb),
+        buf_compute_meta (lctx.buf_compute_meta) {
+            // all initializations should be done in init()
+        }
 
-                    // if output is NULL, init from the input tok embed
-                    if (model.output == NULL) {
-                        model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
-                    }
+    void init() {
+        struct ggml_init_params params = {
+            /*.mem_size   =*/ buf_compute_meta.size(),
+            /*.mem_buffer =*/ buf_compute_meta.data(),
+            /*.no_alloc   =*/ true,
+        };
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        ctx0 = ggml_init(params);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+        lctx.inp_tokens      = nullptr;
+        lctx.inp_embd        = nullptr;
+        lctx.inp_pos         = nullptr;
+        lctx.inp_out_ids     = nullptr;
+        lctx.inp_KQ_mask     = nullptr;
+        lctx.inp_KQ_mask_swa = nullptr;
+        lctx.inp_K_shift     = nullptr;
+        lctx.inp_mean        = nullptr;
+        lctx.inp_cls         = nullptr;
+        lctx.inp_s_copy      = nullptr;
+        lctx.inp_s_mask      = nullptr;
+        lctx.inp_s_seq       = nullptr;
+        lctx.inp_pos_bucket    = nullptr;
+        lctx.inp_embd_enc      = nullptr;
+        lctx.inp_KQ_mask_cross = nullptr;
+    }
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd_head_k * n_head}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_v_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0);
+    void free() {
+        ggml_free(ctx0);
+        ctx0 = nullptr;
+    }
 
-                        // optional bias tensors
-                        layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "bias", i), {n_embd},     llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd},     llama_model_loader::TENSOR_NOT_REQUIRED);
+    struct ggml_cgraph * build_k_shift() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+        GGML_ASSERT(kv_self.size == n_ctx);
 
-                        if (hparams.rope_scaling_type_train == LLAMA_ROPE_SCALING_TYPE_LONGROPE) {
-                            layer.rope_long  = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG,  "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0));
-                            layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0));
-                        }
-                        else {
-                            layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0));
-                        }
+        lctx.inp_K_shift = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_ctx);
+        cb(lctx.inp_K_shift, "K_shift", -1);
+        ggml_set_input(lctx.inp_K_shift);
 
-                        if (n_expert == 0) {
-                            layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                            layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                            layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
+        for (int il = 0; il < n_layer; ++il) {
+            const int64_t n_head_kv = hparams.n_head_kv(il);
+            const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(il);
+            struct ggml_tensor * rope_factors = build_rope_factors(il);
+            struct ggml_tensor * k =
+                ggml_view_3d(ctx0, kv_self.k_l[il],
+                    n_embd_head_k, n_head_kv, n_ctx,
+                    ggml_row_size(kv_self.k_l[il]->type, n_embd_head_k),
+                    ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa),
+                    0);
 
-                            // optional MLP bias
-                            layer.ffn_gate_b = create_tensor(tn(LLM_TENSOR_FFN_GATE, "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                            layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                            layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        } else {
-                            layer.ffn_gate_inp  = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP,  "weight", i), {n_embd, n_expert}, 0);
-                            layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd,   n_ff, n_expert}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                            layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {  n_ff, n_embd, n_expert}, 0);
-                            layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {n_embd,   n_ff, n_expert}, 0);
-                        }
+            struct ggml_tensor * tmp;
+            if (ggml_is_quantized(k->type)) {
+                // dequantize to f32 -> RoPE -> quantize back
+                tmp = ggml_cast(ctx0, k, GGML_TYPE_F32);
+                cb(tmp, "K_f32", il);
+                for (auto & backend : lctx.backends) {
+                    // Figure out which backend KV cache belongs to
+                    if (ggml_backend_supports_buft(backend.get(), ggml_backend_buffer_get_type(kv_self.k_l[il]->buffer))) {
+                        ggml_backend_sched_set_tensor_backend(lctx.sched.get(), tmp, backend.get());
+                        break;
                     }
-                } break;
-            case LLM_ARCH_MINICPM3:
-                {
-                    const int64_t n_embd_head_qk_rope = hparams.n_rot;
-                    const int64_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot;
+                }
+                tmp = ggml_rope_ext_inplace(ctx0, tmp,
+                        lctx.inp_K_shift, rope_factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow);
+                cb(tmp, "K_shifted_f32", il);
+                tmp = ggml_cpy(ctx0, tmp, k);
+            } else {
+                // we rotate only the first n_rot dimensions
+                tmp = ggml_rope_ext_inplace(ctx0, k,
+                        lctx.inp_K_shift, rope_factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow);
+            }
+            cb(tmp, "K_shifted", il);
+            ggml_build_forward_expand(gf, tmp);
+        }
 
-                    const int64_t q_lora_rank  = hparams.n_lora_q;
-                    const int64_t kv_lora_rank = hparams.n_lora_kv;
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        return gf;
+    }
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
+    struct ggml_cgraph * build_defrag(const std::vector & ids) {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                    // if output is NULL, init from the input tok embed
-                    if (model.output == NULL) {
-                        model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
-                    }
+        for (uint32_t i = 0; i < ids.size(); ++i) {
+            const uint32_t id = ids[i];
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            if (i == id || id == ids.size()) {
+                continue;
+            }
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_q_a_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_A_NORM, "weight", i), {q_lora_rank}, 0);
+            uint32_t nm = 1;
 
-                        layer.attn_kv_a_norm = create_tensor(tn(LLM_TENSOR_ATTN_KV_A_NORM, "weight", i), {kv_lora_rank}, 0);
+            while (i + nm < ids.size() && ids[i + nm] == id + nm) {
+                nm++;
+            }
 
-                        layer.wq_a = create_tensor(tn(LLM_TENSOR_ATTN_Q_A, "weight", i), {n_embd, q_lora_rank}, 0);
-                        layer.wq_b = create_tensor(tn(LLM_TENSOR_ATTN_Q_B, "weight", i), {q_lora_rank, n_head * n_embd_head_k}, 0);
+            for (int il = 0; il < n_layer; ++il) {
+                const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(il);
+                const int64_t n_embd_v_gqa = hparams.n_embd_v_gqa(il);
 
-                        layer.wkv_a_mqa = create_tensor(tn(LLM_TENSOR_ATTN_KV_A_MQA, "weight", i), {n_embd, kv_lora_rank + (n_embd_head_qk_rope)}, 0);
-                        layer.wkv_b     = create_tensor(tn(LLM_TENSOR_ATTN_KV_B,     "weight", i), {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)}, 0);
-                        layer.wo        = create_tensor(tn(LLM_TENSOR_ATTN_OUT,      "weight", i), {              n_head * (                      n_embd_head_v), n_embd}, 0);
+                ggml_tensor * view_k_src = ggml_view_2d(ctx0, kv_self.k_l[il],
+                        n_embd_k_gqa, nm,
+                        ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa),
+                        ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa*i));
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+                ggml_tensor * view_k_dst = ggml_view_2d(ctx0, kv_self.k_l[il],
+                        n_embd_k_gqa, nm,
+                        ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa),
+                        ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa*id));
 
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
+                ggml_tensor * view_v_src;
+                ggml_tensor * view_v_dst;
 
-                        layer.rope_long  = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG,  "weight", i), { n_embd_head_qk_rope/2 }, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0));
-                        layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), { n_embd_head_qk_rope/2 }, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0));
-                    }
-                } break;
-            case LLM_ARCH_GROK:
-                {
-                    if (n_expert == 0) {
-                        throw std::runtime_error("Grok model cannot have zero experts");
-                    }
+                if (flash_attn) {
+                    // NOTE: the V cache is not transposed when using flash attention
+                    view_v_src = ggml_view_2d(ctx0, kv_self.v_l[il],
+                            n_embd_v_gqa, nm,
+                            ggml_row_size(kv_self.v_l[il]->type, n_embd_v_gqa),
+                            ggml_row_size(kv_self.v_l[il]->type, n_embd_v_gqa*i));
 
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+                    view_v_dst = ggml_view_2d(ctx0, kv_self.v_l[il],
+                            n_embd_v_gqa, nm,
+                            ggml_row_size(kv_self.v_l[il]->type, n_embd_v_gqa),
+                            ggml_row_size(kv_self.v_l[il]->type, n_embd_v_gqa*id));
+                } else {
+                    view_v_src = ggml_view_2d(ctx0, kv_self.v_l[il],
+                            nm, n_embd_v_gqa,
+                            ggml_row_size(kv_self.v_l[il]->type, kv_self.size),
+                            ggml_row_size(kv_self.v_l[il]->type, i));
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
+                    view_v_dst = ggml_view_2d(ctx0, kv_self.v_l[il],
+                            nm, n_embd_v_gqa,
+                            ggml_row_size(kv_self.v_l[il]->type, kv_self.size),
+                            ggml_row_size(kv_self.v_l[il]->type, id));
+                }
 
-                    // if output is NULL, init from the input tok embed
-                    if (model.output == NULL) {
-                        model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
-                    }
+                ggml_build_forward_expand(gf, ggml_cpy(ctx0, view_k_src, view_k_dst));
+                ggml_build_forward_expand(gf, ggml_cpy(ctx0, view_v_src, view_v_dst));
+            }
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            i += nm - 1;
+        }
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+        //LLAMA_LOG_INFO("gf->n_nodes = %d\n", gf->n_nodes);
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+        return gf;
+    }
 
-                        layer.attn_out_norm   = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}, 0);
+    struct ggml_tensor * build_inp_pos() {
+        lctx.inp_pos = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_tokens);
+        cb(lctx.inp_pos, "inp_pos", -1);
+        ggml_set_input(lctx.inp_pos);
+        return lctx.inp_pos;
+    }
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+    struct ggml_tensor * build_rope_factors(int il) {
+        // choose long/short freq factors based on the context size
+        const auto n_ctx_pre_seq = cparams.n_ctx / cparams.n_seq_max;
 
-                        layer.ffn_gate_inp  = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP,  "weight", i), {n_embd, n_expert}, 0);
-                        layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff, n_expert}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {  n_ff, n_embd, n_expert}, 0);
-                        layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {n_embd,   n_ff, n_expert}, 0);
+        if (model.layers[il].rope_freqs != nullptr) {
+            return model.layers[il].rope_freqs;
+        }
 
-                        layer.layer_out_norm   = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, 0);
-                    }
-                } break;
-            case LLM_ARCH_DBRX:
-                {
-                    if (n_expert == 0) {
-                        throw std::runtime_error("DBRX model cannot have zero experts");
-                    }
+        if (n_ctx_pre_seq > hparams.n_ctx_orig_yarn) {
+            return model.layers[il].rope_long;
+        }
 
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        return model.layers[il].rope_short;
+    }
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+    struct ggml_tensor * build_inp_out_ids() {
+        lctx.inp_out_ids = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_outputs);
+        cb(lctx.inp_out_ids, "inp_out_ids", -1);
+        ggml_set_input(lctx.inp_out_ids);
+        return lctx.inp_out_ids;
+    }
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+    struct ggml_tensor * build_inp_KQ_mask(bool causal = true) {
+        lctx.inp_KQ_mask = causal
+            ? ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv,     GGML_PAD(n_tokens, GGML_KQ_MASK_PAD))
+            : ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_tokens, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
+        cb(lctx.inp_KQ_mask, "KQ_mask", -1);
+        ggml_set_input(lctx.inp_KQ_mask);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+        return flash_attn ? ggml_cast(ctx0, lctx.inp_KQ_mask, GGML_TYPE_F16) : lctx.inp_KQ_mask;
+    }
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
-                        layer.wo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+    struct ggml_tensor * build_inp_KQ_mask_swa(bool causal = true) {
+        GGML_ASSERT(hparams.n_swa > 0);
 
-                        layer.attn_out_norm = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}, 0);
+        lctx.inp_KQ_mask_swa = causal
+            ? ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv,     GGML_PAD(n_tokens, GGML_KQ_MASK_PAD))
+            : ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_tokens, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
+        cb(lctx.inp_KQ_mask_swa, "KQ_mask_swa", -1);
+        ggml_set_input(lctx.inp_KQ_mask_swa);
 
-                        layer.ffn_gate_inp  = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP,  "weight", i), {n_embd, n_expert}, 0);
-                        layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff,   n_expert}, 0);
-                        layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff,   n_embd, n_expert}, 0);
-                        layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {n_embd, n_ff,   n_expert}, 0);
-                    }
-                } break;
-            case LLM_ARCH_BAICHUAN:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
-                    {
-                        model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                        model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
-                    }
+        return flash_attn ? ggml_cast(ctx0, lctx.inp_KQ_mask_swa, GGML_TYPE_F16) : lctx.inp_KQ_mask_swa;
+    }
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+    struct ggml_tensor * build_inp_mean() {
+        lctx.inp_mean = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_tokens, n_tokens);
+        cb(lctx.inp_mean, "inp_mean", -1);
+        ggml_set_input(lctx.inp_mean);
+        return lctx.inp_mean;
+    }
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+    struct ggml_tensor * build_inp_cls() {
+        lctx.inp_cls = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_tokens);
+        cb(lctx.inp_cls, "inp_cls", -1);
+        ggml_set_input(lctx.inp_cls);
+        return lctx.inp_cls;
+    }
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+    struct ggml_tensor * build_inp_s_copy() {
+        lctx.inp_s_copy = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_kv);
+        cb(lctx.inp_s_copy, "inp_s_copy", -1);
+        ggml_set_input(lctx.inp_s_copy);
+        return lctx.inp_s_copy;
+    }
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+    struct ggml_tensor * build_inp_s_mask() {
+        lctx.inp_s_mask = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, 1, n_kv);
+        cb(lctx.inp_s_mask, "inp_s_mask", -1);
+        ggml_set_input(lctx.inp_s_mask);
+        return lctx.inp_s_mask;
+    }
 
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_FALCON:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+    struct ggml_cgraph * append_pooling(struct ggml_cgraph * gf) {
+        // find result_norm tensor for input
+        struct ggml_tensor * inp = nullptr;
+        for (int i = ggml_graph_n_nodes(gf) - 1; i >= 0; --i) {
+            inp = ggml_graph_node(gf, i);
+            if (strcmp(inp->name, "result_norm") == 0 || strcmp(inp->name, "result_embd") == 0) {
+                break;
+            } else {
+                inp = nullptr;
+            }
+        }
+        GGML_ASSERT(inp != nullptr && "missing result_norm/result_embd tensor");
 
-                    // output
-                    {
-                        model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                        model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd}, 0);
+        struct ggml_tensor * cur;
 
-                        model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        if (!model.output) {
-                            model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); // needs to be on GPU
-                        }
-                    }
-
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        switch (pooling_type) {
+            case LLAMA_POOLING_TYPE_NONE:
+                {
+                    cur = inp;
+                } break;
+            case LLAMA_POOLING_TYPE_MEAN:
+                {
+                    struct ggml_tensor * inp_mean = build_inp_mean();
+                    cur = ggml_mul_mat(ctx0, ggml_cont(ctx0, ggml_transpose(ctx0, inp)), inp_mean);
+                } break;
+            case LLAMA_POOLING_TYPE_CLS:
+            case LLAMA_POOLING_TYPE_LAST:
+                {
+                    struct ggml_tensor * inp_cls = build_inp_cls();
+                    cur = ggml_get_rows(ctx0, inp, inp_cls);
+                } break;
+            case LLAMA_POOLING_TYPE_RANK:
+                {
+                    struct ggml_tensor * inp_cls = build_inp_cls();
+                    inp = ggml_get_rows(ctx0, inp, inp_cls);
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i),   {n_embd}, 0);
+                    // classification head
+                    // https://github.com/huggingface/transformers/blob/5af7d41e49bbfc8319f462eb45253dcb3863dfb7/src/transformers/models/roberta/modeling_roberta.py#L1566
+                    GGML_ASSERT(model.cls       != nullptr);
+                    GGML_ASSERT(model.cls_b     != nullptr);
 
-                        layer.attn_norm_2   = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.attn_norm_2_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "bias", i),   {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
+                    cur = ggml_add (ctx0, ggml_mul_mat(ctx0, model.cls, inp), model.cls_b);
+                    cur = ggml_tanh(ctx0, cur);
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
-                        layer.wo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+                    // some models don't have `cls_out`, for example: https://huggingface.co/jinaai/jina-reranker-v1-tiny-en
+                    // https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/blob/cb5347e43979c3084a890e3f99491952603ae1b7/modeling_bert.py#L884-L896
+                    if (model.cls_out) {
+                        GGML_ASSERT(model.cls_out_b != nullptr);
 
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
+                        cur = ggml_add (ctx0, ggml_mul_mat(ctx0, model.cls_out, cur), model.cls_out_b);
                     }
                 } break;
-            case LLM_ARCH_STARCODER:
+            default:
                 {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
-                    model.pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD,   "weight"), {n_embd, n_ctx_train}, 0);
+                    GGML_ABORT("unknown pooling type");
+                }
+        }
 
-                    // output
-                    {
-                        model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                        model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd}, 0);
-                        model.output        = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        if (!model.output) {
-                            // needs to be on GPU
-                            model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
-                        }
+        cb(cur, "result_embd_pooled", -1);
 
-                    }
+        ggml_build_forward_expand(gf, cur);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        return gf;
+    }
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i),   {n_embd}, 0);
+    struct ggml_tensor * llm_build_pos_bucket(bool causal) {
+        if (causal) {
+            lctx.inp_pos_bucket = ggml_new_tensor_2d(ctx0, GGML_TYPE_I32, n_kv,     n_tokens);
+        } else {
+            lctx.inp_pos_bucket = ggml_new_tensor_2d(ctx0, GGML_TYPE_I32, n_tokens, n_tokens);
+        }
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
-                        layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i),   {n_embd + 2*n_embd_gqa}, 0);
+        ggml_set_input(lctx.inp_pos_bucket);
+        cb(lctx.inp_pos_bucket, "pos_bucket", -1);
 
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i),   {n_embd}, 0);
+        return lctx.inp_pos_bucket;
+    }
 
-                        layer.ffn_norm   = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i),   {n_embd}, 0);
+    struct ggml_tensor * llm_build_pos_bias(struct ggml_tensor * pos_bucket, struct ggml_tensor * attn_rel_b) {
+        struct ggml_tensor * pos_bucket_1d = ggml_view_1d(ctx0, pos_bucket, pos_bucket->ne[0] * pos_bucket->ne[1], 0);
+        cb(pos_bucket_1d, "pos_bucket_1d", -1);
 
-                        layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
-                        layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i),   {n_embd}, 0);
+        struct ggml_tensor * pos_bias = ggml_get_rows(ctx0, attn_rel_b, pos_bucket_1d);
+        cb(pos_bias, "pos_bias", -1);
 
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i),   {n_embd, n_ff}, 0);
-                        layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i),     {n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_BERT:
-            case LLM_ARCH_NOMIC_BERT:
-                {
-                    model.tok_embd     = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD,  "weight"), {n_embd, n_vocab}, 0);
-                    model.type_embd    = create_tensor(tn(LLM_TENSOR_TOKEN_TYPES, "weight"), {n_embd, n_vocab_type}, 0);
+        pos_bias = ggml_view_3d(ctx0, pos_bias, pos_bias->ne[0], lctx.inp_pos_bucket->ne[0], lctx.inp_pos_bucket->ne[1], ggml_element_size(pos_bias) * pos_bias->ne[0], ggml_element_size(pos_bias) * pos_bias->ne[0] * lctx.inp_pos_bucket->ne[0],  0);
+        cb(pos_bias, "pos_bias", -1);
 
-                    if (model.arch == LLM_ARCH_BERT) {
-                        model.pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD,    "weight"), {n_embd, n_ctx_train}, 0);
+        pos_bias = ggml_permute(ctx0, pos_bias, 2, 0, 1, 3);
+        cb(pos_bias, "pos_bias", -1);
 
-                        model.cls   = create_tensor(tn(LLM_TENSOR_CLS, "weight"), {n_embd, n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        model.cls_b = create_tensor(tn(LLM_TENSOR_CLS, "bias"),   {n_embd},         llama_model_loader::TENSOR_NOT_REQUIRED);
+        pos_bias = ggml_cont(ctx0, pos_bias);
+        cb(pos_bias, "pos_bias", -1);
 
-                        model.cls_out   = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), {n_embd, 1}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        model.cls_out_b = create_tensor(tn(LLM_TENSOR_CLS_OUT, "bias"),   {1},         llama_model_loader::TENSOR_NOT_REQUIRED);
-                    }
+        return pos_bias;
+    }
 
-                    model.tok_norm   = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight"), {n_embd}, 0);
-                    model.tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias"),   {n_embd}, 0);
+    struct ggml_tensor * llm_build_inp_embd_enc() {
+        const int64_t n_embd = hparams.n_embd;
+        lctx.inp_embd_enc = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_embd, n_outputs_enc);
+        ggml_set_input(lctx.inp_embd_enc);
+        cb(lctx.inp_embd_enc, "embd_enc", -1);
+        return lctx.inp_embd_enc;
+    }
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+    struct ggml_tensor * llm_build_inp_KQ_mask_cross() {
+        lctx.inp_KQ_mask_cross = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_outputs_enc, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
+        ggml_set_input(lctx.inp_KQ_mask_cross);
+        cb(lctx.inp_KQ_mask_cross, "KQ_mask_cross", -1);
+        return lctx.inp_KQ_mask_cross;
+    }
 
-                        if (model.arch == LLM_ARCH_BERT) {
-                            layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                            layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "bias", i),   {n_embd}, 0);
+    struct ggml_cgraph * build_llama() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                            layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                            layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "bias", i),   {n_embd_gqa}, 0);
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
 
-                            layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                            layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "bias", i),   {n_embd_gqa}, 0);
-                        } else {
-                            layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
-                        }
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT,      "weight", i), {n_embd, n_embd}, 0);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-                        layer.attn_out_norm   = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_out_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "bias", i),   {n_embd}, 0);
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,        "weight", i), {n_embd, n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN,      "weight", i), {n_ff, n_embd}, 0);
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-                        if (model.arch == LLM_ARCH_BERT) {
-                            layer.bo         = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0);
-                            layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "bias", i), {n_ff}, 0);
-                            layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, 0);
-                        } else {
-                            layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);
-                        }
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-                        layer.layer_out_norm   = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, 0);
-                        layer.layer_out_norm_b = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "bias", i),   {n_embd}, 0);
-                    }
-                } break;
-            case LLM_ARCH_JINA_BERT_V2:
-                {
-                    model.tok_embd  = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD,  "weight"), {n_embd, n_vocab}, 0); // word_embeddings
-                    model.type_embd = create_tensor(tn(LLM_TENSOR_TOKEN_TYPES, "weight"), {n_embd, n_vocab_type}, 0); // token_type_embeddings
+        const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
 
-                    model.tok_norm   = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight"), {n_embd}, 0); // LayerNorm
-                    model.tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias"),   {n_embd}, 0); //LayerNorm bias
+            // norm
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm", il);
 
-                    model.cls   = create_tensor(tn(LLM_TENSOR_CLS, "weight"), {n_embd, 1}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                    model.cls_b = create_tensor(tn(LLM_TENSOR_CLS, "bias"),   {1},         llama_model_loader::TENSOR_NOT_REQUIRED);
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i]; // JinaBertLayer
+            // self-attention
+            {
+                // rope freq factors for llama3; may return nullptr for llama2 and other models
+                struct ggml_tensor * rope_factors = build_rope_factors(il);
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0);
-                        layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i),   {n_embd}, 0);
+                // compute Q and K and RoPE them
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                cb(Qcur, "Qcur", il);
+                if (model.layers[il].bq) {
+                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                    cb(Qcur, "Qcur", il);
+                }
 
-                        layer.attn_q_norm   = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.attn_q_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "bias",   i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
+                if (model.layers[il].bk) {
+                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                    cb(Kcur, "Kcur", il);
+                }
 
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias",   i), {n_embd_gqa}, 0);
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
+                if (model.layers[il].bv) {
+                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                    cb(Vcur, "Vcur", il);
+                }
 
-                        layer.attn_k_norm   = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.attn_k_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "bias",   i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
+                Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias",   i), {n_embd_gqa}, 0);
+                Kcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0); //output_dens
-                        layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias",   i), {n_embd}, 0); //output_dens
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, model.layers[il].bo,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, kq_scale, cb, il);
+            }
 
-                        layer.attn_out_norm   = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}, 0); //output_norm
-                        layer.attn_out_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "bias",   i), {n_embd}, 0);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                n_tokens = n_outputs;
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            }
 
-                        layer.attn_norm_2   = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.attn_norm_2_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "bias",   i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
+            // For Granite architecture
+            if (hparams.f_residual_scale) {
+                cur = ggml_scale(ctx0, cur, hparams.f_residual_scale);
+            }
 
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_ff}, 0);
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);
-
-                        layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
-                        layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias",   i), {n_embd}, 0);
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
-                        layer.layer_out_norm   = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, 0);
-                        layer.layer_out_norm_b = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "bias",   i), {n_embd}, 0);
-                    }
-                } break;
-            case LLM_ARCH_BLOOM:
-                {
-                    model.tok_embd   = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD,      "weight"), {n_embd, n_vocab}, 0);
-                    model.tok_norm   = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight"), {n_embd}, 0);
-                    model.tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias"),   {n_embd}, 0);
+            // feed-forward network
+            if (model.layers[il].ffn_gate_inp == nullptr) {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "ffn_norm", il);
 
-                    // output
-                    model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd}, 0);
-                    model.output        = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                        model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        NULL,
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                cb(cur, "ffn_out", il);
+            } else {
+                // MoE branch
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "ffn_norm", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+                cur = llm_build_moe_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_gate_inp,
+                        model.layers[il].ffn_up_exps,
+                        model.layers[il].ffn_gate_exps,
+                        model.layers[il].ffn_down_exps,
+                        nullptr,
+                        n_expert, n_expert_used,
+                        LLM_FFN_SILU, true,
+                        false, 0.0,
+                        LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                        cb, il);
+                cb(cur, "ffn_moe_out", il);
+            }
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias",   i), {n_embd}, 0);
+            // For Granite architecture
+            if (hparams.f_residual_scale) {
+                cur = ggml_scale(ctx0, cur, hparams.f_residual_scale);
+            }
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
-                        layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias",   i), {n_embd + 2*n_embd_gqa}, 0);
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cb(cur, "ffn_out", il);
 
-                        layer.wo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.bo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias",   i), {n_embd}, 0);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-                        layer.ffn_norm   = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias",   i), {n_embd}, 0);
+            // input for next layer
+            inpL = cur;
+        }
 
-                        layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
-                        layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias",   i), {n_embd}, 0);
+        cur = inpL;
 
-                        layer.ffn_up     = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0);
-                        layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias",   i), {n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_MPT:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
-                    model.pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD,   "weight"), {n_embd, n_ctx_train}, llama_model_loader::TENSOR_NOT_REQUIRED);
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
+        cb(cur, "result_norm", -1);
 
-                    // output
-                    model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
 
-                    model.output        = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                    if (!model.output) {
-                        model.output    = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); // needs to be on GPU
-                    }
+        // For Granite architecture
+        if (hparams.f_logit_scale) {
+            cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_logit_scale);
+        }
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        cb(cur, "result_output", -1);
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i),   {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
+        ggml_build_forward_expand(gf, cur);
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
-                        layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i),   {n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
+        return gf;
+    }
 
-                        layer.wo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.bo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i),   {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
+    struct ggml_cgraph * build_deci() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                        layer.ffn_norm   = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i),   {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
 
-                        layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
-                        layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i),   {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-                        layer.ffn_up     = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "bias", i),   {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-                        layer.attn_q_norm   = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.attn_q_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "bias",   i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-                        layer.attn_k_norm   = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.attn_k_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "bias",   i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-                        // AWQ ScaleActivation layer
-                        layer.ffn_act = create_tensor(tn(LLM_TENSOR_FFN_ACT, "scales", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                    }
-                } break;
-            case LLM_ARCH_STABLELM:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-                    // output
-                    model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd}, 0);
-                    model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output        = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+        const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
+            const int64_t n_head_kv = hparams.n_head_kv(il);
+            const int64_t n_head    = hparams.n_head(il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            if (n_head == 0) {
+                // attention-free layer of Llama-3_1-Nemotron-51B
+                cur = inpL;
+            } else {
+                // norm
+                cur = llm_build_norm(ctx0, inpL, hparams,
+                        model.layers[il].attn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "attn_norm", il);
+            }
 
-                        layer.attn_norm =   create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0);
+            if (n_head > 0 && n_head_kv == 0) {
+                // "linear attention" of Llama-3_1-Nemotron-51B
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo, cur);
+                cb(cur, "wo", il);
+            } else if (n_head > 0) {
+                // self-attention
+                // rope freq factors for llama3; may return nullptr for llama2 and other models
+                struct ggml_tensor * rope_factors = build_rope_factors(il);
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+                // compute Q and K and RoPE them
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                cb(Qcur, "Qcur", il);
+                if (model.layers[il].bq) {
+                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                    cb(Qcur, "Qcur", il);
+                }
 
-                        // optional bias tensors, present in Stable LM 2 1.6B
-                        layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "bias", i), {n_embd},     llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
+                if (model.layers[il].bk) {
+                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                    cb(Kcur, "Kcur", il);
+                }
 
-                        // optional q and k layernorms, present in StableLM 2 12B
-                        layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k, n_head},    llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k, n_head_kv}, llama_model_loader::TENSOR_NOT_REQUIRED);
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
+                if (model.layers[il].bv) {
+                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                    cb(Vcur, "Vcur", il);
+                }
 
-                        // optional FFN norm, not present in StableLM 2 12B which uses parallel residual
-                        layer.ffn_norm   = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i),   {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
+                Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_QWEN:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+                Kcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, model.layers[il].bo,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, kq_scale, cb, il);
+            }
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                n_tokens = n_outputs;
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            }
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+            // For Granite architecture
+            if (hparams.f_residual_scale) {
+                cur = ggml_scale(ctx0, cur, hparams.f_residual_scale);
+            }
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd*3}, 0);
-                        layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i),   {n_embd*3}, 0);
-                        layer.wo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+            // modified to support attention-free layer of Llama-3_1-Nemotron-51B
+            struct ggml_tensor * ffn_inp = cur;
+            if (n_head > 0) {
+                ffn_inp = ggml_add(ctx0, cur, inpSA);
+                cb(ffn_inp, "ffn_inp", il);
+            }
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+            // feed-forward network
+            if (model.layers[il].ffn_gate_inp == nullptr) {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "ffn_norm", il);
 
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff/2}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff/2, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_ff/2}, 0);
-                    }
-                } break;
-            case LLM_ARCH_QWEN2:
-            case LLM_ARCH_QWEN2VL:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                        model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        NULL,
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                cb(cur, "ffn_out", il);
+            }
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                    // if output is NULL, init from the input tok embed
-                    if (model.output == NULL) {
-                        model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
-                    }
+            // For Granite architecture
+            if (hparams.f_residual_scale) {
+                cur = ggml_scale(ctx0, cur, hparams.f_residual_scale);
+            }
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cb(cur, "ffn_out", il);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+            // input for next layer
+            inpL = cur;
+        }
 
-                        // optional bias tensors
-                        layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "bias", i), {n_embd}, 0);
-                        layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "bias", i), {n_embd_gqa}, 0);
-                        layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "bias", i), {n_embd_gqa}, 0);
+        cur = inpL;
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
+        cb(cur, "result_norm", -1);
 
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_QWEN2MOE:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+        // For Granite architecture
+        if (hparams.f_logit_scale) {
+            cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_logit_scale);
+        }
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        cb(cur, "result_output", -1);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+        ggml_build_forward_expand(gf, cur);
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+        return gf;
+    }
 
-                        // optional bias tensors
-                        layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "bias", i), {n_embd}, 0);
-                        layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "bias", i), {n_embd_gqa}, 0);
-                        layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "bias", i), {n_embd_gqa}, 0);
+    struct ggml_cgraph * build_baichuan() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-                        layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-                        if (n_expert == 0) {
-                            throw std::runtime_error("n_expert must be > 0 for QWEN2MOE");
-                        }
-                        if (n_expert_used == 0) {
-                            throw std::runtime_error("n_expert_used must be > 0 for QWEN2MOE");
-                        }
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-                        // MoE branch
-                        const int64_t n_ff_exp = hparams.n_ff_exp ? hparams.n_ff_exp : n_ff / n_expert_used;
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = model.type == MODEL_7B ? build_inp_pos() : nullptr;
 
-                        layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {  n_embd, n_ff_exp, n_expert}, 0);
-                        layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp,   n_embd, n_expert}, 0);
-                        layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {  n_embd, n_ff_exp, n_expert}, 0);
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-                        // Shared expert branch
-                        const int64_t n_ff_shexp = hparams.n_ff_shexp ? hparams.n_ff_shexp : n_ff;
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
 
-                        layer.ffn_gate_inp_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP_SHEXP, "weight", i), {n_embd}, 0);
-                        layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {    n_embd, n_ff_shexp}, 0);
-                        layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), {n_ff_shexp,     n_embd}, 0);
-                        layer.ffn_up_shexp   = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP,   "weight", i), {    n_embd, n_ff_shexp}, 0);
-                    }
-                } break;
-            case LLM_ARCH_PHI2:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm", il);
 
-                    // output
-                    model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd}, 0);
-                    model.output        = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
-                    model.output_b      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "bias"),   {n_vocab}, 0);
+            // self-attention
+            {
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                cb(Qcur, "Qcur", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i),   {n_embd}, 0);
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i),   {n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
+                switch (model.type) {
+                    case MODEL_7B:
+                        Qcur = ggml_rope_ext(
+                            ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                            n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                            ext_factor, attn_factor, beta_fast, beta_slow
+                        );
+                        Kcur = ggml_rope_ext(
+                            ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                            n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                            ext_factor, attn_factor, beta_fast, beta_slow
+                        );
+                        break;
+                    case MODEL_13B:
+                        Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd/n_head, n_head, n_tokens);
+                        Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd/n_head, n_head, n_tokens);
+                        break;
+                    default:
+                        GGML_ABORT("fatal error");
+                }
+                cb(Qcur, "Qcur", il);
+                cb(Kcur, "Kcur", il);
 
-                        if (layer.wqkv == nullptr) {
-                            layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0);
-                            layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i),   {n_embd}, 0);
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, NULL,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+            }
 
-                            layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0);
-                            layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i),   {n_embd_gqa}, 0);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            }
 
-                            layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0);
-                            layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i),   {n_embd_gqa}, 0);
-                        }
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
-                        layer.wo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.bo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i),   {n_embd}, 0);
+            // feed-forward network
+            {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "ffn_norm", il);
 
-                        layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
-                        layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i),   {n_embd}, 0);
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        model.layers[il].ffn_gate, NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
+                        NULL,
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                cb(cur, "ffn_out", il);
+            }
 
-                        layer.ffn_up     = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_ff}, 0);
-                        layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "bias", i),   {n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_PHI3:
-                {
-                    const int64_t n_embd_head = n_embd / n_head;
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), { n_embd, n_vocab }, 0);
+            // input for next layer
+            inpL = cur;
+        }
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), { n_embd }, 0);
-                    model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), { n_embd, n_vocab }, 0);
+        cur = inpL;
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
+        cb(cur, "result_norm", -1);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), { n_embd }, 0);
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), { n_embd, n_embd + 2 * n_embd_gqa }, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.wo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), { n_embd, n_embd }, 0);
+        ggml_build_forward_expand(gf, cur);
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), { n_embd }, 0);
+        return gf;
+    }
 
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd }, 0);
-                        layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), { n_embd, 2 * n_ff }, 0);
+    struct ggml_cgraph * build_xverse() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                        layer.rope_long  = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG,  "weight", i), { n_embd_head/2 }, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0));
-                        layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), { n_embd_head/2 }, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0));
-                    }
-                } break;
-            case LLM_ARCH_PLAMO:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_GPT2:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
-                    model.pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD,   "weight"), {n_embd, n_ctx_train}, 0);
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-                    // output
-                    model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd}, 0);
-                    model.output        = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm", il);
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM,   "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM,   "bias", i),   {n_embd}, 0);
+            // self-attention
+            {
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                cb(Qcur, "Qcur", il);
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
-                        layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i),   {n_embd + 2*n_embd_gqa}, 0);
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
 
-                        layer.wo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.bo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i),   {n_embd}, 0);
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
 
-                        layer.ffn_norm   = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i),   {n_embd}, 0);
+                Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-                        layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
-                        layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i),   {n_embd}, 0);
+                Kcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, NULL,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+            }
 
-                        layer.ffn_up     = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_ff}, 0);
-                        layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "bias", i),   {n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_CODESHELL:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur   = ggml_get_rows(ctx0,      cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            }
 
-                    // output
-                    model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd}, 0);
-                    model.output        = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            // feed-forward network
+            {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "ffn_norm", il);
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i),   {n_embd}, 0);
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        model.layers[il].ffn_gate, NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
+                        NULL,
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                cb(cur, "ffn_out", il);
+            }
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
-                        layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i),   {n_embd + 2*n_embd_gqa}, 0);
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-                        layer.wo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.bo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i),   {n_embd}, 0);
+            // input for next layer
+            inpL = cur;
+        }
 
-                        layer.ffn_norm   = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i),   {n_embd}, 0);
+        cur = inpL;
 
-                        layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
-                        layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i),   {n_embd}, 0);
+        cur = llm_build_norm(ctx0, cur, hparams, model.output_norm, NULL, LLM_NORM_RMS, cb, -1);
+        cb(cur, "result_norm", -1);
 
-                        layer.ffn_up     = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i),   {n_embd, n_ff}, 0);
-                        layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i),     {n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_ORION:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-                    model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd}, 0);
-                    model.output        = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+        ggml_build_forward_expand(gf, cur);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        return gf;
+    }
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i),   {n_embd}, 0);
+    struct ggml_cgraph * build_falcon() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-                        layer.ffn_norm   = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i),   {n_embd}, 0);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_INTERNLM2:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        // layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * attn_norm;
 
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_GEMMA:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+            attn_norm = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm,
+                    model.layers[il].attn_norm_b,
+                    LLM_NORM, cb, il);
+            cb(attn_norm, "attn_norm", il);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD,  "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); // same as tok_embd, duplicated to allow offloading
+            // self-attention
+            {
+                if (model.layers[il].attn_norm_2) {
+                    // Falcon-40B
+                    cur = llm_build_norm(ctx0, inpL, hparams,
+                            model.layers[il].attn_norm_2,
+                            model.layers[il].attn_norm_2_b,
+                            LLM_NORM, cb, il);
+                    cb(cur, "attn_norm_2", il);
+                } else {
+                    cur = attn_norm;
+                }
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd_head_k * n_head}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_v_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0);
+                cb(Qcur, "Qcur", il);
+                cb(Kcur, "Kcur", il);
+                cb(Vcur, "Vcur", il);
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                    }
-                } break;
-            case LLM_ARCH_GEMMA2:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD,  "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED); // same as tok_embd, duplicated to allow offloading
+                // using mode = 2 for neox mode
+                Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig,
+                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+                Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig,
+                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, NULL,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+            }
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd_head_k * n_head}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_v_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0);
-                        layer.attn_post_norm = create_tensor(tn(LLM_TENSOR_ATTN_POST_NORM, "weight", i), {n_embd}, 0);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur       = ggml_get_rows(ctx0,       cur, inp_out_ids);
+                inpL      = ggml_get_rows(ctx0,      inpL, inp_out_ids);
+                attn_norm = ggml_get_rows(ctx0, attn_norm, inp_out_ids);
+            }
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_post_norm = create_tensor(tn(LLM_TENSOR_FFN_POST_NORM, "weight", i), {n_embd}, 0);
-                    }
-                } break;
-            case LLM_ARCH_STARCODER2:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+            struct ggml_tensor * ffn_inp = cur;
 
-                    // output
-                    model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd}, 0);
+            // feed forward
+            {
+                cur = llm_build_ffn(ctx0, lctx, attn_norm, // !! use the attn norm, not the result
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        NULL,                      NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
+                        NULL,
+                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
+                cb(cur, "ffn_out", il);
+            }
 
-                    model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                    // if output is NULL, init from the input tok embed
-                    if (model.output == NULL) {
-                        model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
-                    }
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = ggml_add(ctx0, cur, inpL);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            // input for next layer
+            inpL = cur;
+        }
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i),   {n_embd}, 0);
+        cur = inpL;
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+        // norm
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm,
+                model.output_norm_b,
+                LLM_NORM, cb, -1);
+        cb(cur, "result_norm", -1);
 
-                        // optional bias tensors
-                        layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "bias", i), {n_embd}, 0);
-                        layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "bias", i), {n_embd_gqa}, 0);
-                        layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "bias", i), {n_embd_gqa}, 0);
-                        layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, 0);
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-                        layer.ffn_norm   = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i),   {n_embd}, 0);
+        ggml_build_forward_expand(gf, cur);
 
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
+        return gf;
+    }
 
-                        // optional bias tensors
-                        layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, 0);
-                        layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP ,  "bias", i), {  n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_MAMBA:
-                {
-                    const int64_t d_conv  = hparams.ssm_d_conv;
-                    const int64_t d_inner = hparams.ssm_d_inner;
-                    const int64_t d_state = hparams.ssm_d_state;
-                    const int64_t dt_rank = hparams.ssm_dt_rank;
+    struct ggml_cgraph * build_grok() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                    // only an expansion factor of 2 is supported for now
-                    if (2 * n_embd != d_inner) {
-                        throw std::runtime_error("only an expansion factor of 2 is supported for now");
-                    }
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
 
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-                    model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                    // if output is NULL, init from the input tok embed, duplicated to allow offloading
-                    if (model.output == NULL) {
-                        model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
-                    }
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        // multiply by embedding_multiplier_scale of 78.38367176906169
+        inpL = ggml_scale(ctx0, inpL, 78.38367176906169f);
 
-                        // norm
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-                        layer.ssm_in = create_tensor(tn(LLM_TENSOR_SSM_IN, "weight", i), {n_embd, 2*d_inner}, 0);
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-                        layer.ssm_conv1d = create_tensor(tn(LLM_TENSOR_SSM_CONV1D, "weight", i), {d_conv, d_inner}, 0);
-                        layer.ssm_conv1d_b = create_tensor(tn(LLM_TENSOR_SSM_CONV1D, "bias", i), {d_inner}, 0);
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
 
-                        layer.ssm_x = create_tensor(tn(LLM_TENSOR_SSM_X, "weight", i), {d_inner, dt_rank + 2*d_state}, 0);
+            // norm
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm", il);
 
-                        layer.ssm_dt = create_tensor(tn(LLM_TENSOR_SSM_DT, "weight", i), {dt_rank, d_inner}, 0);
-                        layer.ssm_dt_b = create_tensor(tn(LLM_TENSOR_SSM_DT, "bias", i), {d_inner}, 0);
 
-                        // no "weight" suffix for these
-                        layer.ssm_a = create_tensor(tn(LLM_TENSOR_SSM_A, i), {d_state, d_inner}, 0);
-                        layer.ssm_d = create_tensor(tn(LLM_TENSOR_SSM_D, i), {d_inner}, 0);
+            // self-attention
+            {
+                // compute Q and K and RoPE them
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                cb(Qcur, "Qcur", il);
+                if (model.layers[il].bq) {
+                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                    cb(Qcur, "Qcur", il);
+                }
 
-                        // out_proj
-                        layer.ssm_out = create_tensor(tn(LLM_TENSOR_SSM_OUT, "weight", i), {d_inner, n_embd}, 0);
-                    }
-                } break;
-            case LLM_ARCH_XVERSE:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
+                if (model.layers[il].bk) {
+                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                    cb(Kcur, "Kcur", il);
+                }
 
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
+                if (model.layers[il].bv) {
+                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                    cb(Vcur, "Vcur", il);
+                }
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+                Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+                Kcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, model.layers[il].bo,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f, cb, il);
+            }
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_COMMAND_R:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                n_tokens = n_outputs;
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            }
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    // init output from the input tok embed
-                    model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
+            // Grok
+            // if attn_out_norm is present then apply it before adding the input
+            if (model.layers[il].attn_out_norm) {
+                cur = llm_build_norm(ctx0, cur, hparams,
+                        model.layers[il].attn_out_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "attn_out_norm", il);
+            }
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+            // feed-forward network
+            // MoE branch
+            cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_norm", il);
 
-                        if (n_layer >= 64){
-                            layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k, n_head}, 0);
-                            layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k, n_head_kv}, 0);
-                        }
+            cur = llm_build_moe_ffn(ctx0, lctx, cur,
+                    model.layers[il].ffn_gate_inp,
+                    model.layers[il].ffn_up_exps,
+                    model.layers[il].ffn_gate_exps,
+                    model.layers[il].ffn_down_exps,
+                    nullptr,
+                    n_expert, n_expert_used,
+                    LLM_FFN_GELU, true,
+                    false, 0.0,
+                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                    cb, il);
+            cb(cur, "ffn_moe_out", il);
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+            // Grok
+            // if layer_out_norm is present then apply it before adding the input
+            // Idea: maybe ffn_out_norm is a better name
+            if (model.layers[il].layer_out_norm) {
+                cur = llm_build_norm(ctx0, cur, hparams,
+                        model.layers[il].layer_out_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "layer_out_norm", il);
+            }
 
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_OLMO:  // adapted from LLM_ARCH_LLAMA with norm params removed
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cb(cur, "ffn_out", il);
 
-                    // output
-                    model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                    // if output is NULL, init from the input tok embed
-                    if (model.output == NULL) {
-                        model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
-                    }
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            // input for next layer
+            inpL = cur;
+        }
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+        cur = inpL;
 
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_OLMO2:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
+        cb(cur, "result_norm", -1);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        // Grok
+        // multiply logits by output_multiplier_scale of 0.5773502691896257
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_post_norm = create_tensor(tn(LLM_TENSOR_ATTN_POST_NORM, "weight", i), {n_embd}, 0);
+        cur = ggml_scale(ctx0, cur, 0.5773502691896257f);
 
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_post_norm = create_tensor(tn(LLM_TENSOR_FFN_POST_NORM, "weight", i), {n_embd}, 0);
-                    }
-                } break;
-            case LLM_ARCH_OLMOE:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        cb(cur, "result_output", -1);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+        ggml_build_forward_expand(gf, cur);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        return gf;
+    }
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+    struct ggml_cgraph * build_dbrx() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd}, 0);
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-                        layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-                        if (n_expert == 0) {
-                            throw std::runtime_error("n_expert must be > 0");
-                        }
-                        if (n_expert_used == 0) {
-                            throw std::runtime_error("n_expert_used must be > 0");
-                        }
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-                        // MoE branch
-                        layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff,   n_expert}, 0);
-                        layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff,   n_embd, n_expert}, 0);
-                        layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {n_embd, n_ff,   n_expert}, 0);
-                    }
-                } break;
-            case LLM_ARCH_OPENELM:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    // init output from the input tok embed
-                    model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        const int64_t n_head      =   hparams.n_head(i);
-                        const int64_t n_head_qkv  = 2*hparams.n_head_kv(i) + n_head;
-                        const int64_t n_ff        =   hparams.n_ff(i);
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
 
-                        auto & layer = model.layers[i];
+            // norm
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                                 model.layers[il].attn_norm, NULL,
+                                 LLM_NORM, cb, il);
+            cb(cur, "attn_norm", il);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+            // self-attention
+            {
+                struct ggml_tensor * Qcur = nullptr;
+                struct ggml_tensor * Kcur = nullptr;
+                struct ggml_tensor * Vcur = nullptr;
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_head_qkv*n_embd_head_k}, 0);
-                        layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k}, 0);
-                        layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_head*n_embd_head_k, n_embd}, 0);
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_GPTNEOX:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+                cur = ggml_clamp(ctx0, cur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
+                cb(cur, "wqkv_clamped", il);
 
-                    // output
-                    model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd}, 0);
-                    model.output        = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+                Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
+                Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
+                Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+                cb(Qcur, "Qcur", il);
+                cb(Kcur, "Kcur", il);
+                cb(Vcur, "Vcur", il);
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i),   {n_embd}, 0);
+                Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
-                        layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i),   {n_embd + 2*n_embd_gqa}, 0);
+                Kcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
-                        layer.wo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.bo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i),   {n_embd}, 0);
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, NULL,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+            }
 
-                        layer.ffn_norm   = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i),   {n_embd}, 0);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                n_tokens = n_outputs;
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            }
 
-                        layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
-                        layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i),   {n_embd}, 0);
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
-                        layer.ffn_up     = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_ff}, 0);
-                        layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "bias", i),   {n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_ARCTIC:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+            // feed-forward network
+            // MoE branch
+            cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                                 model.layers[il].attn_out_norm, NULL,
+                                 LLM_NORM, cb, il);
+            cb(cur, "attn_out_norm", il);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
+            cur = llm_build_moe_ffn(ctx0, lctx, cur,
+                    model.layers[il].ffn_gate_inp,
+                    model.layers[il].ffn_up_exps,
+                    model.layers[il].ffn_gate_exps,
+                    model.layers[il].ffn_down_exps,
+                    nullptr,
+                    n_expert, n_expert_used,
+                    LLM_FFN_SILU, true,
+                    false, 0.0,
+                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                    cb, il);
+            cb(cur, "ffn_moe_out", il);
 
-                    // if output is NULL, init from the input tok embed
-                    if (model.output == NULL) {
-                        model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
-                    }
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cb(cur, "ffn_out", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+            // input for next layer
+            inpL = cur;
+        }
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+        cur = inpL;
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+        cur = llm_build_norm(ctx0, cur, hparams,
+                             model.output_norm, NULL,
+                             LLM_NORM, cb, -1);
+        cb(cur, "result_norm", -1);
 
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_embd}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_embd, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_embd}, 0);
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
 
-                        layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
-                        layer.ffn_norm_exps = create_tensor(tn(LLM_TENSOR_FFN_NORM_EXPS, "weight", i), {n_embd}, 0);
-                        layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd,   n_ff, n_expert}, false);
-                        layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {  n_ff, n_embd, n_expert}, 0);
-                        layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {n_embd,   n_ff, n_expert}, 0);
-                    }
-                } break;
-            case LLM_ARCH_DEEPSEEK:
-                {
+        cb(cur, "result_output", -1);
 
-                    const int64_t n_ff_exp        = hparams.n_ff_exp;
-                    const int64_t n_expert_shared = hparams.n_expert_shared;
+        ggml_build_forward_expand(gf, cur);
 
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        return gf;
+    }
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+    struct ggml_cgraph * build_starcoder() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-                        if (i < (int) hparams.n_layer_dense_lead) {
-                            layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                            layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                            layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                        } else {
-                            layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-                            if (n_expert == 0) {
-                                throw std::runtime_error("n_expert must be > 0");
-                            }
-                            if (n_expert_used == 0) {
-                                throw std::runtime_error("n_expert_used must be > 0");
-                            }
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-                            // MoE branch
-                            layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {  n_embd, n_ff_exp, n_expert}, 0);
-                            layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp,   n_embd, n_expert}, 0);
-                            layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {  n_embd, n_ff_exp, n_expert}, 0);
+        struct ggml_tensor * pos = ggml_get_rows(ctx0, model.pos_embd, inp_pos);
+        cb(pos, "pos_embd", -1);
 
-                            // Shared expert branch
-                            layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0);
-                            layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), {        n_ff_exp * n_expert_shared, n_embd}, 0);
-                            layer.ffn_up_shexp   = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP,   "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0);
-                        }
-                    }
-                } break;
-            case LLM_ARCH_DEEPSEEK2:
-                {
-                    const bool is_lite = (hparams.n_layer == 27);
+        inpL = ggml_add(ctx0, inpL, pos);
+        cb(inpL, "inpL", -1);
 
-                    const int64_t n_embd_head_qk_rope = hparams.n_rot;
-                    const int64_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot;
+        for (int il = 0; il < n_layer; ++il) {
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm,
+                    model.layers[il].attn_norm_b,
+                    LLM_NORM, cb, il);
+            cb(cur, "attn_norm", il);
 
-                    const int64_t q_lora_rank  = hparams.n_lora_q;
-                    const int64_t kv_lora_rank = hparams.n_lora_kv;
+            // self-attention
+            {
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-                    const int64_t n_ff_exp        = hparams.n_ff_exp;
-                    const int64_t n_expert_shared = hparams.n_expert_shared;
+                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
+                cb(cur, "bqkv", il);
 
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+                cb(Qcur, "Qcur", il);
+                cb(Kcur, "Kcur", il);
+                cb(Vcur, "Vcur", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        if (!is_lite) {
-                            layer.attn_q_a_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_A_NORM, "weight", i), {q_lora_rank}, 0);
-                        }
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, model.layers[il].bo,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+            }
 
-                        layer.attn_kv_a_norm = create_tensor(tn(LLM_TENSOR_ATTN_KV_A_NORM, "weight", i), {kv_lora_rank}, 0);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+            }
 
-                        if (!is_lite) {
-                            layer.wq_a = create_tensor(tn(LLM_TENSOR_ATTN_Q_A, "weight", i), {n_embd, q_lora_rank}, 0);
-                            layer.wq_b = create_tensor(tn(LLM_TENSOR_ATTN_Q_B, "weight", i), {q_lora_rank, n_head * n_embd_head_k}, 0);
-                        } else {
-                            layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        }
+            // add the input
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
+            cb(ffn_inp, "ffn_inp", il);
 
-                        layer.wkv_a_mqa = create_tensor(tn(LLM_TENSOR_ATTN_KV_A_MQA, "weight", i), {n_embd, kv_lora_rank + (n_embd_head_qk_rope)}, 0);
-                        layer.wkv_b     = create_tensor(tn(LLM_TENSOR_ATTN_KV_B,     "weight", i), {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)}, 0);
-                        layer.wo        = create_tensor(tn(LLM_TENSOR_ATTN_OUT,      "weight", i), {              n_head * (                      n_embd_head_v), n_embd}, 0);
+            // FF
+            {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm,
+                        model.layers[il].ffn_norm_b,
+                        LLM_NORM, cb, il);
+                cb(cur, "ffn_norm", il);
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                        NULL,                      NULL,                        NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        NULL,
+                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
+                cb(cur, "ffn_out", il);
+            }
 
-                        if (i < (int) hparams.n_layer_dense_lead) {
-                            layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                            layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                            layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                        } else {
-                            layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-                            if (n_expert == 0) {
-                                throw std::runtime_error("n_expert must be > 0");
-                            }
-                            if (n_expert_used == 0) {
-                                throw std::runtime_error("n_expert_used must be > 0");
-                            }
+            // input for next layer
+            inpL = cur;
+        }
 
-                            // MoE branch
-                            layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {  n_embd, n_ff_exp, n_expert}, 0);
-                            layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp,   n_embd, n_expert}, 0);
-                            layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {  n_embd, n_ff_exp, n_expert}, 0);
+        cur = llm_build_norm(ctx0, inpL, hparams,
+                model.output_norm,
+                model.output_norm_b,
+                LLM_NORM, cb, -1);
+        cb(cur, "result_norm", -1);
 
-                            // Shared expert branch
-                            layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0);
-                            layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), {        n_ff_exp * n_expert_shared, n_embd}, 0);
-                            layer.ffn_up_shexp   = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP,   "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0);
-                        }
-                    }
-                } break;
-            case LLM_ARCH_BITNET:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
+        ggml_build_forward_expand(gf, cur);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        return gf;
+    }
 
-                        layer.attn_norm     = create_tensor(tn(LLM_TENSOR_ATTN_NORM,     "weight", i), {n_embd}, 0);
-                        layer.attn_sub_norm = create_tensor(tn(LLM_TENSOR_ATTN_SUB_NORM, "weight", i), {n_embd}, 0);
+    struct ggml_cgraph * build_refact() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                        layer.wq       = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wq_scale = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "scale",  i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.wk       = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wk_scale = create_tensor(tn(LLM_TENSOR_ATTN_K,   "scale",  i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.wv       = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv_scale = create_tensor(tn(LLM_TENSOR_ATTN_V,   "scale",  i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.wo       = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.wo_scale = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "scale",  i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-                        layer.ffn_norm     = create_tensor(tn(LLM_TENSOR_FFN_NORM,     "weight", i), {n_embd}, 0);
-                        layer.ffn_sub_norm = create_tensor(tn(LLM_TENSOR_FFN_SUB_NORM, "weight", i), {n_ff}, 0);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-                        layer.ffn_gate       = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);
-                        layer.ffn_gate_scale = create_tensor(tn(LLM_TENSOR_FFN_GATE, "scale",  i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.ffn_down       = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
-                        layer.ffn_down_scale = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "scale",  i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.ffn_up         = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_ff}, 0);
-                        layer.ffn_up_scale   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "scale",  i), {1}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                    }
-                } break;
-            case LLM_ARCH_T5:
-                {
-                    const auto n_rel_attn_bkts = hparams.n_rel_attn_bkts;
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-                    // output
-                    model.output_norm_enc = create_tensor(tn(LLM_TENSOR_ENC_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output_norm     = create_tensor(tn(LLM_TENSOR_DEC_OUTPUT_NORM, "weight"), {n_embd}, 0);
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
 
-                    model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                    // if output is NULL, init from the input tok embed
-                    if (model.output == NULL) {
-                        model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
-                    }
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            // self-attention
+            {
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                cb(Qcur, "Qcur", il);
 
-                        layer.attn_norm_enc  = create_tensor(tn(LLM_TENSOR_ENC_ATTN_NORM,  "weight", i), {n_embd}, 0);
-                        layer.attn_rel_b_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_REL_B, "weight", i), {n_head, n_rel_attn_bkts}, llama_model_loader::TENSOR_NOT_REQUIRED);
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
 
-                        layer.wq_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_Q,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        layer.wk_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_K,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        layer.wv_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_V,   "weight", i), {n_embd, n_embd_v_gqa}, 0);
-                        layer.wo_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_OUT, "weight", i), {n_embd_v_gqa, n_embd}, 0);
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
 
-                        layer.ffn_norm_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_gate_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_GATE, "weight", i), {n_embd,   n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.ffn_down_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up_enc   = create_tensor(tn(LLM_TENSOR_ENC_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
+                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+                cb(Kcur, "Kcur", il);
 
-                        layer.attn_norm  = create_tensor(tn(LLM_TENSOR_DEC_ATTN_NORM,  "weight", i), {n_embd}, 0);
-                        layer.attn_rel_b = create_tensor(tn(LLM_TENSOR_DEC_ATTN_REL_B, "weight", i), {n_head, n_rel_attn_bkts}, llama_model_loader::TENSOR_NOT_REQUIRED);
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+                cb(Qcur, "Qcur", il);
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_DEC_ATTN_Q,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_DEC_ATTN_K,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_DEC_ATTN_V,   "weight", i), {n_embd, n_embd_v_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_DEC_ATTN_OUT, "weight", i), {n_embd_v_gqa, n_embd}, 0);
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, NULL,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+            }
 
-                        layer.attn_norm_cross  = create_tensor(tn(LLM_TENSOR_DEC_CROSS_ATTN_NORM,  "weight", i), {n_embd}, 0);
-                        // this tensor seems to be unused in HF transformers implementation
-                        layer.attn_rel_b_cross = create_tensor(tn(LLM_TENSOR_DEC_CROSS_ATTN_REL_B, "weight", i), {n_head, n_rel_attn_bkts}, llama_model_loader::TENSOR_NOT_REQUIRED);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            }
 
-                        layer.wq_cross = create_tensor(tn(LLM_TENSOR_DEC_CROSS_ATTN_Q,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        layer.wk_cross = create_tensor(tn(LLM_TENSOR_DEC_CROSS_ATTN_K,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        layer.wv_cross = create_tensor(tn(LLM_TENSOR_DEC_CROSS_ATTN_V,   "weight", i), {n_embd, n_embd_v_gqa}, 0);
-                        layer.wo_cross = create_tensor(tn(LLM_TENSOR_DEC_CROSS_ATTN_OUT, "weight", i), {n_embd_v_gqa, n_embd}, 0);
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_DEC_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_DEC_FFN_GATE, "weight", i), {n_embd,   n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_DEC_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_DEC_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_T5ENCODER:
-                {
-                    const auto n_rel_attn_bkts = hparams.n_rel_attn_bkts;
+            // feed-forward network
+            {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "ffn_norm", il);
 
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        model.layers[il].ffn_gate, NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
+                        NULL,
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                cb(cur, "ffn_out", il);
+            }
 
-                    // output
-                    model.output_norm_enc = create_tensor(tn(LLM_TENSOR_ENC_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                    // if output is NULL, init from the input tok embed
-                    if (model.output == NULL) {
-                        model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
-                    }
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            // input for next layer
+            inpL = cur;
+        }
 
-                        layer.attn_norm_enc  = create_tensor(tn(LLM_TENSOR_ENC_ATTN_NORM,  "weight", i), {n_embd}, 0);
-                        layer.attn_rel_b_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_REL_B, "weight", i), {n_head, n_rel_attn_bkts}, llama_model_loader::TENSOR_NOT_REQUIRED);
+        cur = inpL;
 
-                        layer.wq_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_Q,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        layer.wk_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_K,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        layer.wv_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_V,   "weight", i), {n_embd, n_embd_v_gqa}, 0);
-                        layer.wo_enc = create_tensor(tn(LLM_TENSOR_ENC_ATTN_OUT, "weight", i), {n_embd_v_gqa, n_embd}, 0);
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
+        cb(cur, "result_norm", -1);
 
-                        layer.ffn_norm_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_gate_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_GATE, "weight", i), {n_embd,   n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.ffn_down_enc = create_tensor(tn(LLM_TENSOR_ENC_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up_enc   = create_tensor(tn(LLM_TENSOR_ENC_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_JAIS:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-                    // output
-                    model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd}, 0);
-                    model.output        = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+        ggml_build_forward_expand(gf, cur);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        return gf;
+    }
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM,   "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM,   "bias", i),   {n_embd}, 0);
+    struct ggml_cgraph * build_bert() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
-                        layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i),   {n_embd + 2*n_embd_gqa}, 0);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
 
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
-                        layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i),   {n_embd}, 0);
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-                        layer.ffn_norm   = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i),   {n_embd}, 0);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
+        struct ggml_tensor * inp_pos = nullptr;
 
-                        layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
-                        layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i),   {n_embd}, 0);
+        if (model.arch != LLM_ARCH_JINA_BERT_V2) {
+            inp_pos = build_inp_pos();
+        }
 
-                        layer.ffn_gate   = create_tensor(tn(LLM_TENSOR_FFN_GATE,   "weight", i), {n_embd, n_ff}, 0);
-                        layer.ffn_gate_b = create_tensor(tn(LLM_TENSOR_FFN_GATE,   "bias", i),   {n_ff}, 0);
+        // construct input embeddings (token, type, position)
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-                        layer.ffn_up     = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_ff}, 0);
-                        layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "bias", i),   {n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_CHATGLM:
-                {
-                    model.tok_embd   = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD,      "weight"), {n_embd, n_vocab}, 0);
+        // token types are hardcoded to zero ("Sentence A")
+        struct ggml_tensor * type_row0 = ggml_view_1d(ctx0, model.type_embd, n_embd, 0);
+        inpL = ggml_add(ctx0, inpL, type_row0);
+        if (model.arch == LLM_ARCH_BERT) {
+            inpL = ggml_add(ctx0, ggml_get_rows(ctx0, model.pos_embd, inp_pos), inpL);
+        }
+        cb(inpL, "inp_embd", -1);
 
-                    // output
-                    model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output        = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+        // embed layer norm
+        inpL = llm_build_norm(ctx0, inpL, hparams, model.tok_norm, model.tok_norm_b, LLM_NORM, cb, -1);
+        cb(inpL, "inp_norm", -1);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask(false);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+        // iterate layers
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * cur = inpL;
 
-                        layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, 0);
-                        layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i),   {n_embd + 2*n_embd_gqa}, 0);
+            struct ggml_tensor * Qcur;
+            struct ggml_tensor * Kcur;
+            struct ggml_tensor * Vcur;
 
-                        layer.wo   = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+            // self-attention
+            if (model.arch == LLM_ARCH_BERT || model.arch == LLM_ARCH_JINA_BERT_V2) {
+                Qcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur), model.layers[il].bq);
+                cb(Qcur, "Qcur", il);
 
-                        layer.ffn_norm   = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+                if (model.layers[il].attn_q_norm) {
+                    Qcur = llm_build_norm(ctx0, Qcur, hparams,
+                            model.layers[il].attn_q_norm,
+                            model.layers[il].attn_q_norm_b,
+                            LLM_NORM, cb, il);
+                }
 
-                        layer.ffn_up     = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, n_ff * 2}, 0);
+                Kcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur), model.layers[il].bk);
+                cb(Kcur, "Kcur", il);
 
-                        layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
-                    }
-                } break;
-            case LLM_ARCH_NEMOTRON:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+                if (model.layers[il].attn_k_norm) {
+                    Kcur = llm_build_norm(ctx0, Kcur, hparams,
+                            model.layers[il].attn_k_norm,
+                            model.layers[il].attn_k_norm_b,
+                            LLM_NORM, cb, il);
+                }
+                Vcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur), model.layers[il].bv);
+                cb(Vcur, "Vcur", il);
 
-                    // output
-                    model.output_norm   = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0);
-                    model.output        = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0);
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+            } else {
+                // compute Q and K and RoPE them
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+                Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
+                Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
+                Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, 0);
+                cb(Qcur, "Qcur", il);
+                cb(Kcur, "Kcur", il);
+                cb(Vcur, "Vcur", il);
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+                Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-                        // optional bias tensors
-                        layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "bias", i), {n_embd},     llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd},     llama_model_loader::TENSOR_NOT_REQUIRED);
+                Kcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
+            }
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
-                        layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), {n_embd}, 0);
+            struct ggml_tensor * q =                 ggml_permute(ctx0, Qcur, 0, 2, 1, 3);
+            struct ggml_tensor * k = ggml_cont(ctx0, ggml_permute(ctx0, Kcur, 0, 2, 1, 3));
 
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
+            struct ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
+            cb(kq, "kq", il);
 
-                        // optional MLP bias
-                        layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                    }
-                } break;
-            case LLM_ARCH_EXAONE:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+            kq = ggml_soft_max_ext(ctx0, kq, KQ_mask, 1.0f/sqrtf(float(n_embd_head)), hparams.f_max_alibi_bias);
+            cb(kq, "kq_soft_max_ext", il);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);
+            struct ggml_tensor * v = ggml_cont(ctx0, ggml_transpose(ctx0, ggml_reshape_2d(ctx0, Vcur, n_embd_gqa, n_tokens)));
+            cb(v, "v", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            struct ggml_tensor * kqv = ggml_mul_mat(ctx0, ggml_reshape_3d(ctx0, v, n_tokens, n_embd_head, n_head_kv), kq);
+            cb(kqv, "kqv", il);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
+            struct ggml_tensor * kqv_merged = ggml_permute(ctx0, kqv, 0, 2, 1, 3);
+            cb(kqv_merged, "kqv_merged", il);
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd_head_k * n_head}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_k_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_v_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head, n_embd}, 0);
-
-                        layer.ffn_norm   = create_tensor(tn(LLM_TENSOR_FFN_NORM,   "weight", i), {n_embd}, 0);
-                        layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0));
-                        layer.ffn_gate   = create_tensor(tn(LLM_TENSOR_FFN_GATE,   "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN,   "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up     = create_tensor(tn(LLM_TENSOR_FFN_UP,     "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            case LLM_ARCH_RWKV6:
-                {
-                    model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+            cur = ggml_cont_2d(ctx0, kqv_merged, n_embd_gqa, n_tokens);
+            cb(cur, "kqv_merged_cont", il);
 
-                    // Block 0, LN0
-                    model.tok_norm = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight"), {n_embd}, 0);
-                    model.tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias"), {n_embd}, 0);
+            ggml_build_forward_expand(gf, cur);
 
-                    // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0);
-                    model.output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0);
+            cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo, cur);
+            if (model.layers[il].bo) {
+                cb(cur, "kqv_wo", il);
+            }
 
-                    const int time_mix_extra_dim = hparams.time_mix_extra_dim;
-                    const int time_decay_extra_dim = hparams.time_decay_extra_dim;
-                    const int head_size = hparams.wkv_head_size;
-                    const int attn_hidden_size = n_embd;
-                    const int ffn_size = hparams.n_ff_arr[0];
+            if (model.layers[il].bo) {
+                cur = ggml_add(ctx0, cur, model.layers[il].bo);
+            }
+            cb(cur, "kqv_out", il);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+            if (il == n_layer - 1 && pooling_type == LLAMA_POOLING_TYPE_NONE) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+            }
 
-                        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i),   {n_embd}, 0);
+            // re-add the layer input
+            cur = ggml_add(ctx0, cur, inpL);
 
-                        layer.attn_norm_2   = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "weight", i), {n_embd}, 0);
-                        layer.attn_norm_2_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "bias", i),   {n_embd}, 0);
+            // attention layer norm
+            cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].attn_out_norm, model.layers[il].attn_out_norm_b, LLM_NORM, cb, il);
 
-                        layer.time_mix_w1 = create_tensor(tn(LLM_TENSOR_TIME_MIX_W1, "weight", i), {n_embd, time_mix_extra_dim * 5}, 0);
-                        layer.time_mix_w2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_W2, "weight", i), {time_mix_extra_dim, n_embd, 5}, 0);
+            if (model.layers[il].attn_norm_2 != nullptr) {
+                cur = ggml_add(ctx0, cur, inpL); // re-add the layer input
+                cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].attn_norm_2, model.layers[il].attn_norm_2_b, LLM_NORM, cb, il);
+            }
 
-                        layer.time_mix_lerp_x = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_X, "weight", i), {n_embd, 1, 1}, 0);
-                        layer.time_mix_lerp_w = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_W, "weight", i), {n_embd, 1, 1}, 0);
-                        layer.time_mix_lerp_k = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_K, "weight", i), {n_embd, 1, 1}, 0);
-                        layer.time_mix_lerp_v = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_V, "weight", i), {n_embd, 1, 1}, 0);
-                        layer.time_mix_lerp_r = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_R, "weight", i), {n_embd, 1, 1}, 0);
-                        layer.time_mix_lerp_g = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_G, "weight", i), {n_embd, 1, 1}, 0);
+            struct ggml_tensor * ffn_inp = cur;
+            cb(ffn_inp, "ffn_inp", il);
 
-                        layer.time_mix_first = create_tensor(tn(LLM_TENSOR_TIME_MIX_FIRST, "weight", i), {head_size, n_embd / head_size}, 0);
-                        layer.time_mix_decay = create_tensor(tn(LLM_TENSOR_TIME_MIX_DECAY, "weight", i), {n_embd}, 0);
-                        layer.time_mix_decay_w1 = create_tensor(tn(LLM_TENSOR_TIME_MIX_DECAY_W1, "weight", i), {n_embd, time_decay_extra_dim}, 0);
-                        layer.time_mix_decay_w2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_DECAY_W2, "weight", i), {time_decay_extra_dim, attn_hidden_size}, 0);
-                        layer.time_mix_key = create_tensor(tn(LLM_TENSOR_TIME_MIX_KEY, "weight", i), {attn_hidden_size, n_embd}, 0);
-                        layer.time_mix_value = create_tensor(tn(LLM_TENSOR_TIME_MIX_VALUE, "weight", i), {attn_hidden_size, n_embd}, 0);
-                        layer.time_mix_receptance = create_tensor(tn(LLM_TENSOR_TIME_MIX_RECEPTANCE, "weight", i), {attn_hidden_size, n_embd}, 0);
-                        layer.time_mix_gate = create_tensor(tn(LLM_TENSOR_TIME_MIX_GATE, "weight", i), {attn_hidden_size, n_embd}, 0);
+            // feed-forward network
+            if (model.arch == LLM_ARCH_BERT) {
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                        NULL,                      NULL,                        NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        NULL,
+                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
+            } else if (model.arch == LLM_ARCH_JINA_BERT_V2) {
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   NULL,                        NULL,
+                        model.layers[il].ffn_gate, NULL,                        NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        NULL,
+                        LLM_FFN_GELU, LLM_FFN_PAR, cb, il);
+            } else {
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        model.layers[il].ffn_gate, NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
+                        NULL,
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+            }
+            cb(cur, "ffn_out", il);
 
-                        layer.time_mix_ln = create_tensor(tn(LLM_TENSOR_TIME_MIX_LN, "weight", i), {n_embd}, 0);
-                        layer.time_mix_ln_b = create_tensor(tn(LLM_TENSOR_TIME_MIX_LN, "bias", i), {n_embd}, 0);
-                        layer.time_mix_output = create_tensor(tn(LLM_TENSOR_TIME_MIX_OUTPUT, "weight", i), {n_embd, attn_hidden_size}, 0);
+            // attentions bypass the intermediate layer
+            cur = ggml_add(ctx0, cur, ffn_inp);
 
-                        layer.channel_mix_lerp_k = create_tensor(tn(LLM_TENSOR_CHANNEL_MIX_LERP_K, "weight", i), {n_embd, 1, 1}, 0);
-                        layer.channel_mix_lerp_r = create_tensor(tn(LLM_TENSOR_CHANNEL_MIX_LERP_R, "weight", i), {n_embd, 1, 1}, 0);
+            // output layer norm
+            cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].layer_out_norm, model.layers[il].layer_out_norm_b, LLM_NORM, cb, il);
 
-                        layer.channel_mix_key = create_tensor(tn(LLM_TENSOR_CHANNEL_MIX_KEY, "weight", i), {n_embd, ffn_size}, 0);
-                        layer.channel_mix_value = create_tensor(tn(LLM_TENSOR_CHANNEL_MIX_VALUE, "weight", i), {ffn_size, n_embd}, 0);
-                        layer.channel_mix_receptance = create_tensor(tn(LLM_TENSOR_CHANNEL_MIX_RECEPTANCE, "weight", i), {n_embd, n_embd}, 0);
-                    }
+            // input for next layer
+            inpL = cur;
+        }
 
-                } break;
-            case LLM_ARCH_CHAMELEON:
-                {
-                 model.tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
+        cur = inpL;
 
-                 // output
-                    model.output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
-                    model.output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                    // if output is NULL, init from the input tok embed
-                    if (model.output == NULL) {
-                        model.output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
-                    }
+        cb(cur, "result_embd", -1);
 
-                    for (int i = 0; i < n_layer; ++i) {
-                        auto & layer = model.layers[i];
+        ggml_build_forward_expand(gf, cur);
 
-                        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
-                        layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k, n_head}, 0);
-                        layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k, n_head_kv}, 0);
-                        layer.attn_q_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "bias", i),  {n_embd_head_k, n_head}, llama_model_loader::TENSOR_NOT_REQUIRED);
-                        layer.attn_k_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "bias", i),  {n_embd_head_k, n_head_kv}, llama_model_loader::TENSOR_NOT_REQUIRED);
+        return gf;
+    }
 
-                        layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_gqa}, 0);
-                        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
+    struct ggml_cgraph * build_bloom() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-                        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-                        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, 0);
-                        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, 0);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, 0);
-                    }
-                } break;
-            default:
-                throw std::runtime_error("unknown architecture");
-        }
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-        if (n_moved_tensors > 0) {
-            LLAMA_LOG_DEBUG("%s: tensor '%s' (%s) (and %d others) cannot be used with preferred buffer type %s, using %s instead\n",
-                __func__, first_moved_tensor->name, ggml_type_name(first_moved_tensor->type), n_moved_tensors - 1,
-                ggml_backend_buft_name(first_moved_from_buft), ggml_backend_buft_name(first_moved_to_buft));
-        }
-    }
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-    ml.done_getting_tensors();
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-    ml.init_mappings(true, use_mlock ? &model.mlock_mmaps : nullptr);
-    model.mappings.reserve(ml.mappings.size());
+        inpL = llm_build_norm(ctx0, inpL, hparams,
+                model.tok_norm,
+                model.tok_norm_b,
+                LLM_NORM, cb, -1);
+        cb(inpL, "inp_norm", -1);
 
-    // create the backend buffers
-    std::vector> ctx_bufs;
-    ctx_bufs.reserve(ctx_map.size());
+        for (int il = 0; il < n_layer; ++il) {
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm,
+                    model.layers[il].attn_norm_b,
+                    LLM_NORM, cb, il);
+            cb(cur, "attn_norm", il);
 
-    // Ensure we have enough capacity for the maximum backend buffer we will potentially create
-    const size_t n_max_backend_buffer = ctx_map.size() * ml.files.size();
-    model.bufs.reserve(n_max_backend_buffer);
+            // self-attention
+            {
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-    for (auto & it : ctx_map) {
-        ggml_backend_buffer_type_t buft = it.first;
-        ggml_context * ctx              = it.second;
+                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
+                cb(cur, "bqkv", il);
 
-        // skip contexts without tensors
-        if (ggml_get_first_tensor(ctx) == nullptr) {
-            continue;
-        }
+                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
 
-        llama_buf_map bufs;
-        bufs.reserve(n_max_backend_buffer);
+                cb(Qcur, "Qcur", il);
+                cb(Kcur, "Kcur", il);
+                cb(Vcur, "Vcur", il);
 
-        // check if it is possible to use buffer_from_host_ptr with this buffer type
-        ggml_backend_dev_t dev = ggml_backend_buft_get_device(buft);
-        if (!dev) {
-            // FIXME: workaround for CPU backend buft having a NULL device
-            dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
-        }
-        ggml_backend_dev_props props;
-        ggml_backend_dev_get_props(dev, &props);
-        bool buffer_from_host_ptr_supported = props.caps.buffer_from_host_ptr;
-        bool is_default_buft = buft == ggml_backend_dev_buffer_type(dev);
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
 
-        if (ml.use_mmap && use_mmap_buffer && buffer_from_host_ptr_supported && is_default_buft) {
-            for (uint32_t idx = 0; idx < ml.files.size(); idx++) {
-                // only the mmap region containing the tensors in the model is mapped to the backend buffer
-                // this is important for metal with apple silicon: if the entire model could be mapped to a metal buffer, then we could just use metal for all layers
-                // this allows using partial offloading when the model size exceeds the metal buffer size, but not the RAM size
-                void * addr = nullptr;
-                size_t first, last; // NOLINT
-                ml.get_mapping_range(&first, &last, &addr, idx, ctx);
-                if (first >= last) {
-                    continue;
-                }
-                const size_t max_size = ggml_get_max_tensor_size(ctx);
-                ggml_backend_buffer_t buf = ggml_backend_dev_buffer_from_host_ptr(dev, (char *) addr + first, last - first, max_size);
-                if (buf == nullptr) {
-                    throw std::runtime_error(format("unable to allocate %s buffer", ggml_backend_buft_name(buft)));
-                }
-                model.bufs.emplace_back(buf);
-                bufs.emplace(idx, buf);
-            }
-        }
-        else {
-            ggml_backend_buffer_t buf = ggml_backend_alloc_ctx_tensors_from_buft(ctx, buft);
-            if (buf == nullptr) {
-                throw std::runtime_error(format("unable to allocate %s buffer", ggml_backend_buft_name(buft)));
-            }
-            model.bufs.emplace_back(buf);
-            if (use_mlock && ggml_backend_buffer_is_host(buf)) {
-                model.mlock_bufs.emplace_back(new llama_mlock);
-                auto & mlock_buf = model.mlock_bufs.back();
-                mlock_buf->init   (ggml_backend_buffer_get_base(buf));
-                mlock_buf->grow_to(ggml_backend_buffer_get_size(buf));
-            }
-            for (uint32_t idx = 0; idx < ml.files.size(); idx++) {
-                bufs.emplace(idx, buf);
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, model.layers[il].bo,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
             }
-        }
 
-        if (bufs.empty()) {
-            throw std::runtime_error("failed to allocate buffer");
-        }
-
-        for (auto & buf : bufs) {
-            // indicate that this buffer contains weights
-            // this is used by ggml_backend_sched to improve op scheduling: ops that use a weight are preferably scheduled to the backend that contains the weight
-            ggml_backend_buffer_set_usage(buf.second, GGML_BACKEND_BUFFER_USAGE_WEIGHTS);
-        }
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+            }
 
-        ctx_bufs.emplace_back(ctx, bufs);
-    }
+            // Add the input
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
+            cb(ffn_inp, "ffn_inp", il);
 
-    if (llama_supports_gpu_offload()) {
-        const int n_gpu = std::min(n_gpu_layers, int(hparams.n_layer));
+            // FF
+            {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm,
+                        model.layers[il].ffn_norm_b,
+                        LLM_NORM, cb, il);
+                cb(cur, "ffn_norm", il);
 
-        LLAMA_LOG_INFO("%s: offloading %d repeating layers to GPU\n", __func__, n_gpu);
-        if (n_gpu_layers > (int) hparams.n_layer) {
-            LLAMA_LOG_INFO("%s: offloading output layer to GPU\n", __func__);
-        }
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                        NULL,                      NULL,                        NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        NULL,
+                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
+                cb(cur, "ffn_out", il);
+            }
 
-        const int max_backend_supported_layers = hparams.n_layer + 1;
-        const int max_offloadable_layers       = hparams.n_layer + 1;
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-        LLAMA_LOG_INFO("%s: offloaded %d/%d layers to GPU\n", __func__, std::min(n_gpu_layers, max_offloadable_layers), max_backend_supported_layers);
-    }
+            // input for next layer
+            inpL = cur;
+        }
 
-    // print memory requirements per buffer type
-    for (auto & buf : model.bufs) {
-        LLAMA_LOG_INFO("%s: %12s model buffer size = %8.2f MiB\n", __func__, ggml_backend_buffer_name(buf.get()), ggml_backend_buffer_get_size(buf.get()) / 1024.0 / 1024.0);
-    }
+        cur = llm_build_norm(ctx0, inpL, hparams,
+                model.output_norm,
+                model.output_norm_b,
+                LLM_NORM, cb, -1);
+        cb(cur, "result_norm", -1);
 
-    // populate tensors_by_name
-    for (auto & ctx : model.ctxs) {
-        for (auto * cur = ggml_get_first_tensor(ctx.get()); cur != NULL; cur = ggml_get_next_tensor(ctx.get(), cur)) {
-            model.tensors_by_name.emplace_back(ggml_get_name(cur), cur);
-        }
-    }
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-    // load tensor data
-    for (auto & it : ctx_bufs) {
-        ggml_context * ctx = it.first;
-        auto & bufs = it.second;
-        if (!ml.load_all_data(ctx, bufs, use_mlock ? &model.mlock_mmaps : NULL, progress_callback, progress_callback_user_data)) {
-            return false;
-        }
-    }
+        ggml_build_forward_expand(gf, cur);
 
-    if (use_mmap_buffer) {
-        for (auto & mapping : ml.mappings) {
-            model.mappings.emplace_back(std::move(mapping));
-        }
+        return gf;
     }
 
-    return true;
-}
-
-// Returns 0 on success, -1 on error, and -2 on cancellation via llama_progress_callback
-static int llama_model_load(const std::string & fname, llama_model & model, llama_model_params & params) {
-    model.t_start_us = ggml_time_us();
+    struct ggml_cgraph * build_mpt() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-    try {
-        llama_model_loader ml(fname, params.use_mmap, params.check_tensors, params.kv_overrides);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-        model.hparams.vocab_only = params.vocab_only;
+        struct ggml_tensor * cur;
+        struct ggml_tensor * pos;
+        struct ggml_tensor * inpL;
 
-        try {
-            llm_load_arch(ml, model);
-        } catch(const std::exception & e) {
-            throw std::runtime_error("error loading model architecture: " + std::string(e.what()));
-        }
-        try {
-            llm_load_hparams(ml, model);
-        } catch(const std::exception & e) {
-            throw std::runtime_error("error loading model hyperparameters: " + std::string(e.what()));
-        }
-        try {
-            llm_load_vocab(ml, model);
-        } catch(const std::exception & e) {
-            throw std::runtime_error("error loading model vocabulary: " + std::string(e.what()));
-        }
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-        llm_load_stats(ml, model);
-        llm_load_print_meta(ml, model);
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-        if (model.vocab.type != LLAMA_VOCAB_TYPE_NONE &&
-            model.hparams.n_vocab != model.vocab.id_to_token.size()) {
-            throw std::runtime_error("vocab size mismatch");
-        }
+        if (model.pos_embd) {
+            // inp_pos - contains the positions
+            struct ggml_tensor * inp_pos = build_inp_pos();
+            pos = ggml_get_rows(ctx0, model.pos_embd, inp_pos);
+            cb(pos, "pos_embd", -1);
 
-        if (params.vocab_only) {
-            LLAMA_LOG_INFO("%s: vocab only - skipping tensors\n", __func__);
-            return 0;
+            inpL = ggml_add(ctx0, inpL, pos);
+            cb(inpL, "inpL", -1);
         }
 
-        if (!llm_load_tensors(
-            ml, model, params.n_gpu_layers, params.split_mode,  params.main_gpu, params.tensor_split, params.use_mlock,
-            params.progress_callback, params.progress_callback_user_data
-        )) {
-            return -2;
-        }
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: error loading model: %s\n", __func__, err.what());
-        return -1;
-    }
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * attn_norm;
 
-    // loading time will be recalculate after the first eval, so
-    // we take page faults deferred by mmap() into consideration
-    model.t_load_us = ggml_time_us() - model.t_start_us;
+            attn_norm = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm,
+                    model.layers[il].attn_norm_b,
+                    LLM_NORM, cb, il);
+            cb(attn_norm, "attn_norm", il);
 
-    return 0;
-}
+            // self-attention
+            {
+                cur = attn_norm;
 
-//
-// llm_build
-//
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-using llm_build_cb = std::function;
+                if (model.layers[il].bqkv){
+                    cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
+                    cb(cur, "bqkv", il);
+                }
 
-enum llm_ffn_op_type {
-    LLM_FFN_SILU,
-    LLM_FFN_GELU,
-    LLM_FFN_RELU,
-    LLM_FFN_RELU_SQR,
-    LLM_FFN_SWIGLU,
-};
+                if (hparams.f_clamp_kqv > 0.0f) {
+                    cur = ggml_clamp(ctx0, cur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
+                    cb(cur, "wqkv_clamped", il);
+                }
 
-enum llm_ffn_gate_type {
-    LLM_FFN_SEQ,
-    LLM_FFN_PAR, // ffn_gate is parallel to ffn_up
-};
+                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
 
-enum llm_norm_type {
-    LLM_NORM,
-    LLM_NORM_RMS,
-};
+                cb(Qcur, "Qcur", il);
+                cb(Kcur, "Kcur", il);
+                cb(Vcur, "Vcur", il);
 
-static struct ggml_tensor * llm_build_inp_embd(
-        struct ggml_context * ctx,
-       struct llama_context & lctx,
-        const llama_hparams & hparams,
-         const llama_ubatch & batch,
-         struct ggml_tensor * tok_embd,
-         const llm_build_cb & cb) {
-    const int64_t n_embd = hparams.n_embd;
+                // Q/K Layernorm
+                if (model.layers[il].attn_q_norm) {
+                    Qcur = llm_build_norm(ctx0, Qcur, hparams,
+                            model.layers[il].attn_q_norm,
+                            model.layers[il].attn_q_norm_b,
+                            LLM_NORM, cb, il);
+                    cb(Qcur, "Qcur", il);
 
-    struct ggml_tensor * inpL;
+                    Kcur = llm_build_norm(ctx0, Kcur, hparams,
+                            model.layers[il].attn_k_norm,
+                            model.layers[il].attn_k_norm_b,
+                            LLM_NORM, cb, il);
+                    cb(Kcur, "Kcur", il);
 
-    if (batch.token) {
-        lctx.inp_tokens = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, batch.n_tokens);
-        cb(lctx.inp_tokens, "inp_tokens", -1);
-        ggml_set_input(lctx.inp_tokens);
+                    Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+                    Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
 
-        inpL = ggml_get_rows(ctx, tok_embd, lctx.inp_tokens);
-    } else {
-       lctx.inp_embd = ggml_new_tensor_2d(ctx, GGML_TYPE_F32, n_embd, batch.n_tokens);
-        inpL = lctx.inp_embd;
-        ggml_set_input(lctx.inp_embd);
-    }
+                    cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                            model.layers[il].wo, model.layers[il].bo,
+                            Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+                } else {
+                    Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
 
-    // For Granite architecture
-    if (hparams.f_embedding_scale != 0.0f) {
-        inpL = ggml_scale(ctx, inpL, hparams.f_embedding_scale);
-    }
+                    cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                            model.layers[il].wo, model.layers[il].bo,
+                            Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+                }
+            }
 
-    cb(inpL, "inp_embd", -1);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+            }
 
-    return inpL;
-}
+            // Add the input
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
+            cb(ffn_inp, "ffn_inp", il);
 
-static void llm_build_kv_store(
-        struct ggml_context * ctx,
-        const llama_hparams & hparams,
-        const llama_cparams & cparams,
-       const llama_kv_cache & kv,
-         struct ggml_cgraph * graph,
-         struct ggml_tensor * k_cur,
-         struct ggml_tensor * v_cur,
-                    int32_t   n_tokens,
-                    int32_t   kv_head,
-         const llm_build_cb & cb,
-                    int64_t   il) {
-    const int64_t n_ctx = cparams.n_ctx;
-
-    const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(il);
-    const int64_t n_embd_v_gqa = hparams.n_embd_v_gqa(il);
+            // feed forward
+            {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm,
+                        model.layers[il].ffn_norm_b,
+                        LLM_NORM, cb, il);
+                cb(cur, "ffn_norm", il);
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                        NULL,                      NULL,                        NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        model.layers[il].ffn_act,
+                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
+                cb(cur, "ffn_out", il);
+            }
 
-    GGML_ASSERT(kv.size == n_ctx);
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-    struct ggml_tensor * k_cache_view = ggml_view_1d(ctx, kv.k_l[il], n_tokens*n_embd_k_gqa, ggml_row_size(kv.k_l[il]->type, n_embd_k_gqa)*kv_head);
-    cb(k_cache_view, "k_cache_view", il);
+            // input for next layer
+            inpL = cur;
+        }
 
-    // note: storing RoPE-ed version of K in the KV cache
-    ggml_build_forward_expand(graph, ggml_cpy(ctx, k_cur, k_cache_view));
+        cur = inpL;
 
-    assert(v_cur->ne[0] == n_embd_v_gqa && v_cur->ne[1] == n_tokens);
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm,
+                model.output_norm_b,
+                LLM_NORM, cb, -1);
+        cb(cur, "result_norm", -1);
 
-    struct ggml_tensor * v_cache_view = nullptr;
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-    if (cparams.flash_attn) {
-        v_cache_view = ggml_view_1d(ctx, kv.v_l[il], n_tokens*n_embd_v_gqa, ggml_row_size(kv.v_l[il]->type, n_embd_v_gqa)*kv_head);
-    } else {
-        // note: the V cache is transposed when not using flash attention
-        v_cache_view = ggml_view_2d(ctx, kv.v_l[il], n_tokens, n_embd_v_gqa,
-                (  n_ctx)*ggml_element_size(kv.v_l[il]),
-                (kv_head)*ggml_element_size(kv.v_l[il]));
+        ggml_build_forward_expand(gf, cur);
 
-        v_cur = ggml_transpose(ctx, v_cur);
+        return gf;
     }
-    cb(v_cache_view, "v_cache_view", il);
-
-    ggml_build_forward_expand(graph, ggml_cpy(ctx, v_cur, v_cache_view));
-}
 
-// do mat_mul, while optionally apply lora
-static struct ggml_tensor * llm_build_lora_mm(
-        struct llama_context & lctx,
-         struct ggml_context * ctx0,
-          struct ggml_tensor * w,
-          struct ggml_tensor * cur) {
-    struct ggml_tensor * res = ggml_mul_mat(ctx0, w, cur);
-    for (auto & it : lctx.lora_adapters) {
-        struct llama_lora_weight * lora = it.first->get_weight(w);
-        if (lora == nullptr) {
-            continue;
-        }
-        const float alpha = it.first->alpha;
-        const float rank  = (float) lora->b->ne[0];
-        const float scale = alpha ? it.second * alpha / rank : it.second;
-        struct ggml_tensor * ab_cur = ggml_mul_mat(
-            ctx0, lora->b,
-            ggml_mul_mat(ctx0, lora->a, cur)
-        );
-        ab_cur = ggml_scale(ctx0, ab_cur, scale);
-        res = ggml_add(ctx0, res, ab_cur);
-    }
-    return res;
-}
+    struct ggml_cgraph * build_stablelm() {
+        struct ggml_cgraph * gf = ggml_new_graph(ctx0);
 
-// do mat_mul_id, while optionally apply lora
-static struct ggml_tensor * llm_build_lora_mm_id(
-        struct llama_context & lctx,
-         struct ggml_context * ctx0,
-          struct ggml_tensor * w,   // struct ggml_tensor * as
-          struct ggml_tensor * cur, // struct ggml_tensor * b
-          struct ggml_tensor * ids) {
-    struct ggml_tensor * res = ggml_mul_mat_id(ctx0, w, cur, ids);
-    for (auto & it : lctx.lora_adapters) {
-        struct llama_lora_weight * lora = it.first->get_weight(w);
-        if (lora == nullptr) {
-            continue;
-        }
-        const float alpha = it.first->alpha;
-        const float rank  = (float) lora->b->ne[0];
-        const float scale = alpha ? it.second * alpha / rank : it.second;
-        struct ggml_tensor * ab_cur = ggml_mul_mat_id(
-            ctx0, lora->b,
-            ggml_mul_mat_id(ctx0, lora->a, cur, ids),
-            ids
-        );
-        ab_cur = ggml_scale(ctx0, ab_cur, scale);
-        res = ggml_add(ctx0, res, ab_cur);
-    }
-    return res;
-}
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-static struct ggml_tensor * llm_build_norm(
-        struct ggml_context * ctx,
-         struct ggml_tensor * cur,
-        const llama_hparams & hparams,
-         struct ggml_tensor * mw,
-         struct ggml_tensor * mb,
-              llm_norm_type   type,
-         const llm_build_cb & cb,
-                        int   il) {
-    switch (type) {
-        case LLM_NORM:     cur = ggml_norm    (ctx, cur, hparams.f_norm_eps);     break;
-        case LLM_NORM_RMS: cur = ggml_rms_norm(ctx, cur, hparams.f_norm_rms_eps); break;
-    }
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-    if (mw || mb) {
-        cb(cur, "norm", il);
-    }
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-    if (mw) {
-        cur = ggml_mul(ctx, cur, mw);
-        if (mb) {
-            cb(cur, "norm_w", il);
-        }
-    }
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-    if (mb) {
-        cur = ggml_add(ctx, cur, mb);
-    }
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-    return cur;
-}
+        for (int il = 0; il < n_layer; ++il) {
 
-static struct ggml_tensor * llm_build_ffn(
-        struct ggml_context * ctx,
-       struct llama_context & lctx,
-         struct ggml_tensor * cur,
-         struct ggml_tensor * up,
-         struct ggml_tensor * up_b,
-         struct ggml_tensor * up_s,
-         struct ggml_tensor * gate,
-         struct ggml_tensor * gate_b,
-         struct ggml_tensor * gate_s,
-         struct ggml_tensor * down,
-         struct ggml_tensor * down_b,
-         struct ggml_tensor * down_s,
-         struct ggml_tensor * act_scales,
-            llm_ffn_op_type   type_op,
-          llm_ffn_gate_type   type_gate,
-         const llm_build_cb & cb,
-                        int   il) {
-    struct ggml_tensor * tmp = up ? llm_build_lora_mm(lctx, ctx, up, cur) : cur;
-    cb(tmp, "ffn_up", il);
 
-    if (up_b) {
-        tmp = ggml_add(ctx, tmp, up_b);
-        cb(tmp, "ffn_up_b", il);
-    }
+            // norm
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm,
+                    model.layers[il].attn_norm_b,
+                    LLM_NORM, cb, il);
+            cb(cur, "attn_norm", il);
 
-    if (up_s) {
-        tmp = ggml_mul(ctx, tmp, up_s);
-        cb(tmp, "ffn_up_s", il);
-    }
+            struct ggml_tensor * inpSA = cur;
 
-    if (gate) {
-        switch (type_gate) {
-            case LLM_FFN_SEQ:
-                {
-                    cur = llm_build_lora_mm(lctx, ctx, gate, tmp);
-                    cb(cur, "ffn_gate", il);
-                } break;
-            case LLM_FFN_PAR:
-                {
-                    cur = llm_build_lora_mm(lctx, ctx, gate, cur);
-                    cb(cur, "ffn_gate", il);
-                } break;
-        }
+            // self-attention
+            {
+                // compute Q and K and RoPE them
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                cb(Qcur, "Qcur", il);
+                if (model.layers[il].bq) {
+                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                    cb(Qcur, "Qcur", il);
+                }
 
-        if (gate_b) {
-            cur = ggml_add(ctx, cur, gate_b);
-            cb(cur, "ffn_gate_b", il);
-        }
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
+                if (model.layers[il].bk) {
+                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                    cb(Kcur, "Kcur", il);
+                }
 
-        if (gate_s) {
-            cur = ggml_mul(ctx, cur, gate_s);
-            cb(cur, "ffn_gate_s", il);
-        }
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
+                if (model.layers[il].bv) {
+                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                    cb(Vcur, "Vcur", il);
+                }
 
-    } else {
-        cur = tmp;
-    }
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+                cb(Qcur, "Qcur", il);
+                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+                cb(Kcur, "Kcur", il);
 
-    switch (type_op) {
-        case LLM_FFN_SILU:
-            {
-                cur = ggml_silu(ctx, cur);
-                cb(cur, "ffn_silu", il);
-            } break;
-        case LLM_FFN_GELU:
-            {
-                cur = ggml_gelu(ctx, cur);
-                cb(cur, "ffn_gelu", il);
-                if (act_scales != NULL) {
-                    cur = ggml_div(ctx, cur, act_scales);
-                    cb(cur, "ffn_act", il);
+                if (model.layers[il].attn_q_norm) {
+                    Qcur = llm_build_norm(ctx0, Qcur, hparams,
+                            model.layers[il].attn_q_norm,
+                            NULL,
+                            LLM_NORM, cb, il);
+                    cb(Qcur, "Qcur", il);
+                }
+                if (model.layers[il].attn_k_norm) {
+                    Kcur = llm_build_norm(ctx0, Kcur, hparams,
+                            model.layers[il].attn_k_norm,
+                            NULL,
+                            LLM_NORM, cb, il);
+                    cb(Kcur, "Kcur", il);
                 }
-            } break;
-        case LLM_FFN_RELU:
-            {
-                cur = ggml_relu(ctx, cur);
-                cb(cur, "ffn_relu", il);
-            } break;
-        case LLM_FFN_RELU_SQR:
-            {
-                cur = ggml_relu(ctx, cur);
-                cb(cur, "ffn_relu", il);
 
-                cur = ggml_sqr(ctx, cur);
-                cb(cur, "ffn_sqr(relu)", il);
-            } break;
-        case LLM_FFN_SWIGLU:
-            {
-                // Project to 4h. If using swiglu double the output width, see https://arxiv.org/pdf/2002.05202.pdf
-                int64_t split_point = cur->ne[0] / 2;
-                struct ggml_tensor * x0 = ggml_cont(ctx, ggml_view_2d(ctx, cur, split_point, cur->ne[1], cur->nb[1], 0));
-                struct ggml_tensor * x1 = ggml_cont(ctx, ggml_view_2d(ctx, cur, split_point, cur->ne[1], cur->nb[1], split_point * ggml_element_size(cur)));
 
-                x0 = ggml_silu(ctx, x0);
-                cb(cur, "ffn_silu", il);
+                Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-                cur = ggml_mul(ctx, x0, x1);
-                cb(cur, "ffn_mul", il);
-            } break;
-    }
+                Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
-    if (type_gate == LLM_FFN_PAR) {
-        cur = ggml_mul(ctx, cur, tmp);
-        cb(cur, "ffn_gate_par", il);
-    }
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, NULL,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+            }
 
-    if (down) {
-        cur = llm_build_lora_mm(lctx, ctx, down, cur);
-    }
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpL  = ggml_get_rows(ctx0,  inpL, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            }
 
-    if (down_b) {
-        cb(cur, "ffn_down", il);
-    }
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
+            cb(ffn_inp, "ffn_inp", il);
 
-    if (down_b) {
-        cur = ggml_add(ctx, cur, down_b);
-    }
+            // feed-forward network
+            {
+                if (model.layers[il].ffn_norm) {
+                    cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                            model.layers[il].ffn_norm,
+                            model.layers[il].ffn_norm_b,
+                            LLM_NORM, cb, il);
+                    cb(cur, "ffn_norm", il);
+                } else {
+                    // parallel residual
+                    cur = inpSA;
+                }
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        model.layers[il].ffn_gate, NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
+                        NULL,
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                cb(cur, "ffn_out", il);
+            }
 
-    if (down_s) {
-        cur = ggml_mul(ctx, cur, down_s);
-        cb(cur, "ffn_down_s", il);
-    }
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-    return cur;
-}
+            // input for next layer
+            inpL = cur;
+        }
 
-static struct ggml_tensor * llm_build_moe_ffn(
-        struct ggml_context * ctx,
-       struct llama_context & lctx,
-         struct ggml_tensor * cur,
-         struct ggml_tensor * gate_inp,
-         struct ggml_tensor * up_exps,
-         struct ggml_tensor * gate_exps,
-         struct ggml_tensor * down_exps,
-                    int64_t   n_expert,
-                    int64_t   n_expert_used,
-            llm_ffn_op_type   type_op,
-                       bool   norm_w,
-                       bool   scale_w,
-                      float   w_scale,
-         const llm_build_cb & cb,
-                        int   il) {
-    int64_t n_embd = cur->ne[0];
-    int64_t n_tokens = cur->ne[1];
+        cur = inpL;
 
-    ggml_tensor * logits = llm_build_lora_mm(lctx, ctx, gate_inp, cur); // [n_expert, n_tokens]
-    cb(logits, "ffn_moe_logits", il);
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm,
+                model.output_norm_b,
+                LLM_NORM, cb, -1);
+        cb(cur, "result_norm", -1);
 
-    ggml_tensor * probs = ggml_soft_max(ctx, logits); // [n_expert, n_tokens]
-    cb(probs, "ffn_moe_probs", il);
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-    // select experts
-    ggml_tensor * selected_experts = ggml_top_k(ctx, probs, n_expert_used); // [n_expert_used, n_tokens]
-    cb(selected_experts->src[0], "ffn_moe_argsort", il);
-    cb(selected_experts, "ffn_moe_topk", il);
+        ggml_build_forward_expand(gf, cur);
 
-    ggml_tensor * weights = ggml_get_rows(ctx,
-            ggml_reshape_3d(ctx, probs, 1, n_expert, n_tokens), selected_experts); // [1, n_expert_used, n_tokens]
-    cb(weights, "ffn_moe_weights", il);
+        return gf;
+    }
 
-    if (norm_w) {
-        weights = ggml_reshape_2d(ctx, weights, n_expert_used, n_tokens);
+    struct ggml_cgraph * build_qwen() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        ggml_tensor * weights_sum = ggml_sum_rows(ctx, weights); // [1, n_tokens]
-        cb(weights_sum, "ffn_moe_weights_sum", il);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-        weights = ggml_div(ctx, weights, weights_sum); // [n_expert_used, n_tokens]
-        cb(weights, "ffn_moe_weights_norm", il);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-        weights = ggml_reshape_3d(ctx, weights, 1, n_expert_used, n_tokens);
-    }
-    if (scale_w) {
-        weights = ggml_scale(ctx, weights, w_scale);
-        cb(weights, "ffn_moe_weights_scaled", il);
-    }
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-    cur = ggml_reshape_3d(ctx, cur, n_embd, 1, n_tokens);
-    ggml_tensor * up = llm_build_lora_mm_id(lctx, ctx, up_exps, cur, selected_experts); // [n_ff, n_expert_used, n_tokens]
-    cb(up, "ffn_moe_up", il);
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-    ggml_tensor * gate = llm_build_lora_mm_id(lctx, ctx, gate_exps, cur, selected_experts); // [n_ff, n_expert_used, n_tokens]
-    cb(gate, "ffn_moe_gate", il);
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-    switch (type_op) {
-        case LLM_FFN_SILU:
-            {
-                gate = ggml_silu(ctx, gate);
-                cb(gate, "ffn_moe_silu", il);
-            } break;
-        case LLM_FFN_GELU:
-            {
-                gate = ggml_gelu(ctx, gate);
-                cb(gate, "ffn_moe_gelu", il);
-            } break;
-        default:
-            GGML_ABORT("fatal error");
-    }
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
 
-    ggml_tensor * par = ggml_mul(ctx, up, gate); // [n_ff, n_expert_used, n_tokens]
-    cb(par, "ffn_moe_gate_par", il);
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm", il);
 
-    ggml_tensor * experts = llm_build_lora_mm_id(lctx, ctx, down_exps, par, selected_experts); // [n_embd, n_expert_used, n_tokens]
-    cb(experts, "ffn_moe_down", il);
+            // self-attention
+            {
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-    experts = ggml_mul(ctx, experts, weights);
+                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
+                cb(cur, "bqkv", il);
 
-    // aggregate experts
-    ggml_tensor * moe_out = nullptr;
-    for (int i = 0; i < n_expert_used; ++i) {
-        ggml_tensor * cur_expert = ggml_view_2d(ctx, experts, n_embd, n_tokens,
-                experts->nb[2], i*experts->nb[1]);
+                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd, n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd, n_tokens, cur->nb[1], 2*sizeof(float)*(n_embd)));
 
-        if (i == 0) {
-            moe_out = cur_expert;
-        } else {
-            moe_out = ggml_add(ctx, moe_out, cur_expert);
-        }
-    }
+                cb(Qcur, "Qcur", il);
+                cb(Kcur, "Kcur", il);
+                cb(Vcur, "Vcur", il);
 
-    if (n_expert_used == 1) {
-        // avoid returning a non-contiguous tensor
-        moe_out = ggml_cont(ctx, moe_out);
-    }
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
 
-    return moe_out;
-}
+                // using mode = 2 for neox mode
+                Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig,
+                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-static struct ggml_tensor * llm_build_kqv(
-        struct ggml_context * ctx,
-       struct llama_context & lctx,
-       const llama_kv_cache & kv,
-         struct ggml_cgraph * graph,
-         struct ggml_tensor * wo,
-         struct ggml_tensor * wo_b,
-         struct ggml_tensor * q_cur,
-         struct ggml_tensor * kq_mask,
-                    int32_t   n_tokens,
-                    int32_t   n_kv,
-                    float     kq_scale,
-         const llm_build_cb & cb,
-                    int       il) {
-    const llama_model   & model   = lctx.model;
-    const llama_hparams & hparams = lctx.model.hparams;
-    const llama_cparams & cparams = lctx.cparams;
+                Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig,
+                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
-    const int64_t n_ctx         = cparams.n_ctx;
-    const int64_t n_head        = hparams.n_head(il);
-    const int64_t n_head_kv     = hparams.n_head_kv(il);
-    const int64_t n_embd_head_k = hparams.n_embd_head_k;
-    const int64_t n_embd_k_gqa  = hparams.n_embd_k_gqa(il);
-    const int64_t n_embd_head_v = hparams.n_embd_head_v;
-    const int64_t n_embd_v_gqa  = hparams.n_embd_v_gqa(il);
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, NULL,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+            }
 
-    struct ggml_tensor * q = ggml_permute(ctx, q_cur, 0, 2, 1, 3);
-    cb(q, "q", il);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            }
 
-    struct ggml_tensor * k =
-        ggml_view_3d(ctx, kv.k_l[il],
-                n_embd_head_k, n_kv, n_head_kv,
-                ggml_row_size(kv.k_l[il]->type, n_embd_k_gqa),
-                ggml_row_size(kv.k_l[il]->type, n_embd_head_k),
-                0);
-    cb(k, "k", il);
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
-    struct ggml_tensor * cur;
+            // feed-forward forward
+            {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "ffn_norm", il);
 
-    if (cparams.flash_attn) {
-        GGML_UNUSED(model);
-        GGML_UNUSED(n_ctx);
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        model.layers[il].ffn_gate, NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
+                        NULL,
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                cb(cur, "ffn_out", il);
+            }
 
-        // split cached v into n_head heads (not transposed)
-        struct ggml_tensor * v =
-            ggml_view_3d(ctx, kv.v_l[il],
-                    n_embd_head_v, n_kv, n_head_kv,
-                    ggml_row_size(kv.v_l[il]->type, n_embd_v_gqa),
-                    ggml_row_size(kv.v_l[il]->type, n_embd_head_v),
-                    0);
-        cb(v, "v", il);
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-        cur = ggml_flash_attn_ext(ctx, q, k, v, kq_mask, kq_scale, hparams.f_max_alibi_bias,
-                                  hparams.attn_soft_cap ? hparams.f_attn_logit_softcapping : 0.0f);
+            // input for next layer
+            inpL = cur;
+        }
 
-        ggml_flash_attn_ext_set_prec(cur, GGML_PREC_F32);
+        cur = inpL;
 
-        cur = ggml_reshape_2d(ctx, cur, n_embd_head_v*n_head, n_tokens);
-    } else {
-        struct ggml_tensor * kq = ggml_mul_mat(ctx, k, q);
-        cb(kq, "kq", il);
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
+        cb(cur, "result_norm", -1);
 
-        // note: this op tends to require high floating point range
-        //       while for some models F16 is enough, for others it is not, so we default to F32 here
-        ggml_mul_mat_set_prec(kq, GGML_PREC_F32);
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-        if (model.arch == LLM_ARCH_GROK) {
-            // need to do the following:
-            // multiply by attn_output_multiplyer of 0.08838834764831845
-            // and then :
-            // kq = 30 * tanh(kq / 30)
-            // before the softmax below
+        ggml_build_forward_expand(gf, cur);
 
-            kq = ggml_tanh(ctx, ggml_scale(ctx, kq, 0.08838834764831845f/30.0f));
-            kq = ggml_scale(ctx, kq, 30);
-        }
+        return gf;
+    }
 
-        if (hparams.attn_soft_cap) {
-            kq = ggml_scale(ctx, kq, 1.0f / hparams.f_attn_logit_softcapping);
-            kq = ggml_tanh(ctx, kq);
-            kq = ggml_scale(ctx, kq, hparams.f_attn_logit_softcapping);
-        }
+    struct ggml_cgraph * build_qwen2() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        kq = ggml_soft_max_ext(ctx, kq, kq_mask, kq_scale, hparams.f_max_alibi_bias);
-        cb(kq, "kq_soft_max_ext", il);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-        GGML_ASSERT(kv.size == n_ctx);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-        // split cached v into n_head heads
-        struct ggml_tensor * v =
-            ggml_view_3d(ctx, kv.v_l[il],
-                    n_kv, n_embd_head_v, n_head_kv,
-                    ggml_element_size(kv.v_l[il])*n_ctx,
-                    ggml_element_size(kv.v_l[il])*n_ctx*n_embd_head_v,
-                    0);
-        cb(v, "v", il);
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-        struct ggml_tensor * kqv = ggml_mul_mat(ctx, v, kq);
-        cb(kqv, "kqv", il);
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-        struct ggml_tensor * kqv_merged = ggml_permute(ctx, kqv, 0, 2, 1, 3);
-        cb(kqv_merged, "kqv_merged", il);
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-        cur = ggml_cont_2d(ctx, kqv_merged, n_embd_head_v*n_head, n_tokens);
-        cb(cur, "kqv_merged_cont", il);
-    }
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
 
-    ggml_build_forward_expand(graph, cur);
+            // norm
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm", il);
 
-    if (wo) {
-        cur = llm_build_lora_mm(lctx, ctx, wo, cur);
-    }
+            // self-attention
+            {
+                // compute Q and K and RoPE them
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                cb(Qcur, "Qcur", il);
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                cb(Qcur, "Qcur", il);
 
-    if (wo_b) {
-        cb(cur, "kqv_wo", il);
-    }
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                cb(Kcur, "Kcur", il);
 
-    if (wo_b) {
-        cur = ggml_add(ctx, cur, wo_b);
-    }
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                cb(Vcur, "Vcur", il);
 
-    return cur;
-}
+                Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-static struct ggml_tensor * llm_build_kv(
-        struct ggml_context * ctx,
-       struct llama_context & lctx,
-       const llama_kv_cache & kv,
-         struct ggml_cgraph * graph,
-         struct ggml_tensor * wo,
-         struct ggml_tensor * wo_b,
-         struct ggml_tensor * k_cur,
-         struct ggml_tensor * v_cur,
-         struct ggml_tensor * q_cur,
-         struct ggml_tensor * kq_mask,
-                    int32_t   n_tokens,
-                    int32_t   kv_head,
-                    int32_t   n_kv,
-                    float     kq_scale,
-         const llm_build_cb & cb,
-                    int       il) {
-    const llama_hparams & hparams = lctx.model.hparams;
-    const llama_cparams & cparams = lctx.cparams;
+                Kcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
-    // these nodes are added to the graph together so that they are not reordered
-    // by doing so, the number of splits in the graph is reduced
-    ggml_build_forward_expand(graph, q_cur);
-    ggml_build_forward_expand(graph, k_cur);
-    ggml_build_forward_expand(graph, v_cur);
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, model.layers[il].bo,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+            }
 
-    llm_build_kv_store(ctx, hparams, cparams, kv, graph, k_cur, v_cur, n_tokens, kv_head, cb, il);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            }
 
-    struct ggml_tensor * cur;
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
-    cur  = llm_build_kqv(ctx, lctx, kv, graph, wo, wo_b, q_cur, kq_mask, n_tokens, n_kv, kq_scale, cb, il);
-    cb(cur, "kqv_out", il);
+            // feed-forward network
+            cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_norm", il);
 
-    return cur;
-}
+            cur = llm_build_ffn(ctx0, lctx, cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+            cb(cur, "ffn_out", il);
 
-static struct ggml_tensor * llm_build_copy_mask_state(
-        struct ggml_context * ctx,
-         struct ggml_cgraph * graph,
-         struct ggml_tensor * s,
-         struct ggml_tensor * state_copy,
-         struct ggml_tensor * state_mask,
-                    int32_t   n_state,
-                    int32_t   kv_size,
-                    int32_t   kv_head,
-                    int32_t   n_kv,
-                    int32_t   n_seqs) {
-    struct ggml_tensor * states = ggml_reshape_2d(ctx, s, n_state, kv_size);
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-    // copy states
-    // NOTE: assuming the copy destinations are ALL contained between kv_head and kv_head + n_kv
-    // this shrinks the tensors's ne[1] to n_kv
-    states = ggml_get_rows(ctx, states, state_copy);
+            // input for next layer
+            inpL = cur;
+        }
 
-    // clear states of sequences which are starting at the beginning of this batch
-    // FIXME: zero-out NANs?
-    states = ggml_mul(ctx, states, state_mask);
+        cur = inpL;
 
-    // copy states which won't be changed further (between n_seqs and n_kv)
-    ggml_build_forward_expand(graph,
-        ggml_cpy(ctx,
-            ggml_view_1d(ctx, states, n_state*(n_kv - n_seqs), n_seqs*n_state*ggml_element_size(states)),
-            ggml_view_1d(ctx, s, n_state*(n_kv - n_seqs), (kv_head + n_seqs)*n_state*ggml_element_size(s))));
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
+        cb(cur, "result_norm", -1);
 
-    // the part of the states that will be used and modified
-    return ggml_view_2d(ctx, states, n_state, n_seqs, states->nb[1], 0);
-}
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-// TODO: split
-static struct ggml_tensor * llm_build_mamba(
-        struct ggml_context * ctx,
-       struct llama_context & lctx,
-         const llama_ubatch & batch,
-         struct ggml_cgraph * graph,
-         struct ggml_tensor * cur,
-         struct ggml_tensor * state_copy,
-         struct ggml_tensor * state_mask,
-                    int32_t   kv_head,
-                    int32_t   n_kv,
-         const llm_build_cb & cb,
-                    int       il) {
-    const llama_model    & model   = lctx.model;
-    const llama_hparams  & hparams = model.hparams;
-    const llama_kv_cache & kv      = lctx.kv_self;
-    const int64_t d_conv  = hparams.ssm_d_conv;
-    const int64_t d_inner = hparams.ssm_d_inner;
-    const int64_t d_state = hparams.ssm_d_state;
-    const int64_t dt_rank = hparams.ssm_dt_rank;
-    const int64_t n_seqs  = batch.n_seqs;
-    // Some variants of Mamba arch (e.g. FalconMamba do apply layer norm on B and Dt layers)
-    const bool ssm_dt_b_c_rms = hparams.ssm_dt_b_c_rms;
-    // Use the same RMS norm as the final layer norm
-    const float norm_rms_eps = hparams.f_norm_rms_eps;
+        ggml_build_forward_expand(gf, cur);
 
-    const int64_t n_seq_tokens = batch.n_seq_tokens;
+        return gf;
+    }
 
-    GGML_ASSERT(n_seqs != 0);
-    GGML_ASSERT(batch.equal_seqs);
-    GGML_ASSERT(batch.n_tokens == n_seq_tokens * n_seqs);
+    struct ggml_cgraph * build_qwen2vl() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-    struct ggml_tensor * conv_states_all = kv.k_l[il];
-    struct ggml_tensor * ssm_states_all  = kv.v_l[il];
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-    // (ab)using the KV cache to store the states
-    struct ggml_tensor * conv = llm_build_copy_mask_state(ctx,
-            graph, conv_states_all, state_copy, state_mask,
-            hparams.n_embd_k_s(), kv.size, kv_head, n_kv, n_seqs);
-    conv = ggml_reshape_3d(ctx, conv, d_conv - 1, d_inner, n_seqs);
-    struct ggml_tensor * ssm = llm_build_copy_mask_state(ctx,
-            graph, ssm_states_all, state_copy, state_mask,
-            hparams.n_embd_v_s(), kv.size, kv_head, n_kv, n_seqs);
-    ssm = ggml_reshape_3d(ctx, ssm, d_state, d_inner, n_seqs);
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-    // {n_embd, n_tokens} => {n_embd, n_seq_tokens, n_seqs}
-    cur = ggml_reshape_3d(ctx, cur, cur->ne[0], n_seq_tokens, n_seqs);
+        // inp_pos - contains the positions
+        lctx.inp_pos = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_tokens * 4);
+        cb(lctx.inp_pos, "inp_pos", -1);
+        ggml_set_input(lctx.inp_pos);
+        struct ggml_tensor * inp_pos = lctx.inp_pos;
 
-    // {n_embd, 2*d_inner} @ {n_embd, n_seq_tokens, n_seqs} => {2*d_inner, n_seq_tokens, n_seqs}
-    struct ggml_tensor * xz = llm_build_lora_mm(lctx, ctx, model.layers[il].ssm_in, cur);
-    // split the above in two
-    // => {d_inner, n_seq_tokens, n_seqs}
-    struct ggml_tensor * x = ggml_view_3d(ctx, xz, d_inner, xz->ne[1], xz->ne[2], xz->nb[1], xz->nb[2], 0);
-    struct ggml_tensor * z = ggml_view_3d(ctx, xz, d_inner, xz->ne[1], xz->ne[2], xz->nb[1], xz->nb[2], d_inner*ggml_element_size(xz));
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
+        int sections[4];
+        std::copy(std::begin(hparams.rope_sections), std::begin(hparams.rope_sections) + 4, sections);
 
-    // conv
-    {
-        // => {d_conv - 1 + n_seq_tokens, d_inner, n_seqs}
-        struct ggml_tensor * conv_x = ggml_concat(ctx, conv, ggml_transpose(ctx, x), 0);
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
 
-        // copy last (d_conv - 1) columns back into the state cache
-        struct ggml_tensor * last_conv = ggml_view_3d(ctx, conv_x, d_conv - 1, d_inner, n_seqs, conv_x->nb[1], conv_x->nb[2], n_seq_tokens*(conv_x->nb[0]));
+            // norm
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm", il);
 
-        ggml_build_forward_expand(graph,
-            ggml_cpy(ctx, last_conv,
-                ggml_view_1d(ctx, conv_states_all,
-                    (d_conv - 1)*(d_inner)*(n_seqs),
-                    kv_head*(d_conv - 1)*(d_inner)*ggml_element_size(conv_states_all))));
+            // self-attention
+            {
+                // compute Q and K and RoPE them
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                cb(Qcur, "Qcur", il);
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                cb(Qcur, "Qcur", il);
 
-        // 1D convolution
-        // The equivalent is to make a self-overlapping view of conv_x
-        // over d_conv columns at each stride in the 3rd dimension,
-        // then element-wise multiply that with the conv1d weight,
-        // then sum the elements of each row,
-        // (the last two steps are a dot product over rows (also doable with mul_mat))
-        // then permute away the ne[0] dimension,
-        // and then you're left with the resulting x tensor.
-        // For simultaneous sequences, all sequences need to have the same length.
-        x = ggml_ssm_conv(ctx, conv_x, model.layers[il].ssm_conv1d);
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                cb(Kcur, "Kcur", il);
 
-        // bias
-        x = ggml_add(ctx, x, model.layers[il].ssm_conv1d_b);
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                cb(Vcur, "Vcur", il);
 
-        x = ggml_silu(ctx, x);
-    }
+                Qcur = ggml_rope_multi(
+                    ctx0,
+                    ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                    n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-    // ssm
-    {
-        // {d_inner, dt_rank + 2*d_state} @ {d_inner, n_seq_tokens, n_seqs} => {dt_rank + 2*d_state, n_seq_tokens, n_seqs}
-        struct ggml_tensor * x_db = llm_build_lora_mm(lctx, ctx, model.layers[il].ssm_x, x);
-        // split
-        struct ggml_tensor * dt = ggml_view_3d(ctx, x_db, dt_rank, n_seq_tokens, n_seqs, x_db->nb[1], x_db->nb[2], 0);
-        struct ggml_tensor * B  = ggml_view_3d(ctx, x_db, d_state, n_seq_tokens, n_seqs, x_db->nb[1], x_db->nb[2], ggml_element_size(x_db)*dt_rank);
-        struct ggml_tensor * C  = ggml_view_3d(ctx, x_db, d_state, n_seq_tokens, n_seqs, x_db->nb[1], x_db->nb[2], ggml_element_size(x_db)*(dt_rank+d_state));
+                Kcur = ggml_rope_multi(
+                    ctx0,
+                    ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
-        // Some Mamba variants (e.g. FalconMamba) apply RMS norm in B, C & Dt layers
-        if (ssm_dt_b_c_rms) {
-            dt = ggml_rms_norm(ctx, dt, norm_rms_eps);
-            B = ggml_rms_norm(ctx, B, norm_rms_eps);
-            C = ggml_rms_norm(ctx, C, norm_rms_eps);
-        }
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, model.layers[il].bo,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+            }
 
-        // {dt_rank, d_inner} @ {dt_rank, n_seq_tokens, n_seqs} => {d_inner, n_seq_tokens, n_seqs}
-        dt = llm_build_lora_mm(lctx, ctx, model.layers[il].ssm_dt, dt);
-        dt = ggml_add(ctx, dt, model.layers[il].ssm_dt_b);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            }
 
-        // Custom operator to optimize the parallel associative scan
-        // as described in the Annex D of the Mamba paper.
-        // => {d_inner, n_seq_tokens, n_seqs} and {d_state, d_inner, n_seqs}
-        struct ggml_tensor * y_ssm = ggml_ssm_scan(ctx, ssm, x, dt, model.layers[il].ssm_a, B, C);
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
-        // store last states
-        ggml_build_forward_expand(graph,
-            ggml_cpy(ctx,
-                ggml_view_1d(ctx, y_ssm, d_state*d_inner*n_seqs, x->nb[3]),
-                ggml_view_1d(ctx, ssm_states_all, d_state*d_inner*n_seqs, kv_head*d_state*d_inner*ggml_element_size(ssm_states_all))));
+            // feed-forward network
+            cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_norm", il);
 
-        struct ggml_tensor * y = ggml_view_3d(ctx, y_ssm, d_inner, n_seq_tokens, n_seqs, x->nb[1], x->nb[2], 0);
+            cur = llm_build_ffn(ctx0, lctx, cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+            cb(cur, "ffn_out", il);
 
-        // TODO: skip computing output earlier for unused tokens
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-        // {d_inner, n_seq_tokens, n_seqs} * {d_inner} => {d_inner, n_seq_tokens, n_seqs}
-        y = ggml_add(ctx, y, ggml_mul(ctx, x, model.layers[il].ssm_d));
-        y = ggml_mul(ctx, y, ggml_silu(ctx, ggml_cont(ctx, z)));
+            // input for next layer
+            inpL = cur;
+        }
 
-        // {d_inner, n_embd} @ {d_inner, n_seq_tokens, n_seqs} => {n_embd, n_seq_tokens, n_seqs}
-        cur = llm_build_lora_mm(lctx, ctx, model.layers[il].ssm_out, y);
-    }
+        cur = inpL;
 
-    // {n_embd, n_seq_tokens, n_seqs} => {n_embd, n_tokens}
-    cur = ggml_reshape_2d(ctx, cur, cur->ne[0], n_seq_tokens * n_seqs);
-    cb(cur, "mamba_out", il);
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
+        cb(cur, "result_norm", -1);
 
-    return cur;
-}
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-static struct ggml_tensor * llm_build_rwkv6_time_mix(
-        struct llama_context & lctx,
-        struct ggml_context * ctx,
-        const struct llama_layer * layer,
-        struct ggml_tensor * cur,
-        struct ggml_tensor * x_prev,
-        struct ggml_tensor ** wkv_state) {
-    size_t n_embd       = cur->ne[0];
-    size_t n_seq_tokens = cur->ne[1];
-    size_t n_seqs       = cur->ne[2];
+        ggml_build_forward_expand(gf, cur);
 
-    size_t head_size  = layer->time_mix_first->ne[0];
-    size_t head_count = layer->time_mix_first->ne[1];
+        return gf;
+    }
 
-    size_t n_tokens = n_seqs * n_seq_tokens;
+    struct ggml_cgraph * build_qwen2moe() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-    struct ggml_tensor * sx = ggml_sub(ctx, x_prev, cur);
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
 
-    sx  = ggml_reshape_2d(ctx, sx,  n_embd, n_tokens);
-    cur = ggml_reshape_2d(ctx, cur, n_embd, n_tokens);
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
-    struct ggml_tensor * xxx = ggml_add(ctx, ggml_mul(ctx, sx, layer->time_mix_lerp_x), cur);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-    xxx = ggml_reshape_4d(
-        ctx,
-        ggml_tanh(
-            ctx,
-            ggml_mul_mat(ctx, layer->time_mix_w1, xxx)
-        ),
-        layer->time_mix_w1->ne[1] / 5, 1, 5, n_tokens
-    );
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-    xxx = ggml_cont(ctx, ggml_permute(ctx, xxx, 0, 1, 3, 2));
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-    xxx = ggml_mul_mat(
-        ctx,
-        ggml_reshape_4d(
-            ctx,
-            layer->time_mix_w2,
-            layer->time_mix_w2->ne[0], layer->time_mix_w2->ne[1], 1, 5
-        ),
-        xxx
-    );
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-    struct ggml_tensor *mw = ggml_view_2d(ctx, xxx, n_embd, n_tokens, xxx->nb[1], 0);
-    struct ggml_tensor *mk = ggml_view_2d(ctx, xxx, n_embd, n_tokens, xxx->nb[1], n_embd * n_tokens * sizeof(float));
-    struct ggml_tensor *mv = ggml_view_2d(ctx, xxx, n_embd, n_tokens, xxx->nb[1], n_embd * n_tokens * 2 * sizeof(float));
-    struct ggml_tensor *mr = ggml_view_2d(ctx, xxx, n_embd, n_tokens, xxx->nb[1], n_embd * n_tokens * 3 * sizeof(float));
-    struct ggml_tensor *mg = ggml_view_2d(ctx, xxx, n_embd, n_tokens, xxx->nb[1], n_embd * n_tokens * 4 * sizeof(float));
+        for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
 
-    struct ggml_tensor * xw = ggml_add(
-        ctx,
-        ggml_mul(
-            ctx,
-            ggml_add(ctx, mw, layer->time_mix_lerp_w),
-            sx
-        ),
-        cur
-    );
+            // norm
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm", il);
 
-    struct ggml_tensor * xk = ggml_add(
-        ctx,
-        ggml_mul(
-            ctx,
-            ggml_add(ctx, mk, layer->time_mix_lerp_k),
-            sx
-        ),
-        cur
-    );
+            // self_attention
+            {
+                // compute Q and K and RoPE them
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                cb(Qcur, "Qcur", il);
+                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                cb(Qcur, "Qcur", il);
 
-    struct ggml_tensor * xv = ggml_add(
-        ctx,
-        ggml_mul(
-            ctx,
-            ggml_add(ctx, mv, layer->time_mix_lerp_v),
-            sx
-        ),
-        cur
-    );
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
+                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                cb(Kcur, "Kcur", il);
 
-    struct ggml_tensor * xr = ggml_add(
-        ctx,
-        ggml_mul(
-            ctx,
-            ggml_add(ctx, mr, layer->time_mix_lerp_r),
-            sx
-        ),
-        cur
-    );
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
+                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                cb(Vcur, "Vcur", il);
 
-    struct ggml_tensor * xg = ggml_add(
-        ctx,
-        ggml_mul(
-            ctx,
-            ggml_add(ctx, mg, layer->time_mix_lerp_g),
-            sx
-        ),
-        cur
-    );
+                Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-    struct ggml_tensor * r = ggml_reshape_4d(ctx, llm_build_lora_mm(lctx, ctx, layer->time_mix_receptance, xr), head_size, 1,         head_count, n_tokens);
-    struct ggml_tensor * k = ggml_reshape_4d(ctx, llm_build_lora_mm(lctx, ctx, layer->time_mix_key,        xk), 1,         head_size, head_count, n_tokens);
-    struct ggml_tensor * v = ggml_reshape_4d(ctx, llm_build_lora_mm(lctx, ctx, layer->time_mix_value,      xv), head_size, 1,         head_count, n_tokens);
-    struct ggml_tensor * g = ggml_silu(
-        ctx,
-        llm_build_lora_mm(lctx, ctx, layer->time_mix_gate, xg)
-    );
+                Kcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
-    struct ggml_tensor * w = ggml_mul_mat(
-        ctx,
-        layer->time_mix_decay_w2,
-        ggml_tanh(
-            ctx,
-            ggml_mul_mat(ctx, layer->time_mix_decay_w1, xw)
-        )
-    );
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, model.layers[il].bo,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+            }
 
-    w = ggml_add(ctx, w, ggml_reshape_1d(ctx, layer->time_mix_decay, n_embd));
-    w = ggml_exp(ctx, ggml_neg(ctx, ggml_exp(ctx, w)));
-    w = ggml_reshape_4d(ctx, w, 1, head_size, head_count, n_tokens);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                n_tokens = n_outputs;
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+            }
 
-    k = ggml_transpose(ctx, k);
-    v = ggml_transpose(ctx, v);
-    r = ggml_transpose(ctx, r);
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
-    struct ggml_tensor * wkv_output = ggml_rwkv_wkv6(ctx, k, v, r, layer->time_mix_first, w, *wkv_state);
-    cur = ggml_view_1d(ctx, wkv_output, n_embd * n_tokens, 0);
-    *wkv_state = ggml_view_1d(ctx, wkv_output, n_embd * head_size * n_seqs, n_embd * n_tokens * sizeof(float));
+            // MoE branch
+            cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_norm", il);
 
-    // group norm with head_count groups
-    cur = ggml_reshape_3d(ctx, cur, n_embd / head_count, head_count, n_tokens);
-    cur = ggml_norm(ctx, cur, 64e-5f);
+            ggml_tensor * moe_out =
+                    llm_build_moe_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_gate_inp,
+                        model.layers[il].ffn_up_exps,
+                        model.layers[il].ffn_gate_exps,
+                        model.layers[il].ffn_down_exps,
+                        nullptr,
+                        n_expert, n_expert_used,
+                        LLM_FFN_SILU, false,
+                        false, 0.0,
+                        LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                        cb, il);
+            cb(cur, "ffn_moe_out", il);
 
-    // Convert back to regular vectors.
-    cur = ggml_reshape_2d(ctx, cur, n_embd, n_tokens);
-    cur = ggml_add(ctx, ggml_mul(ctx, cur, layer->time_mix_ln), layer->time_mix_ln_b);
+            // FFN shared expert
+            {
+                ggml_tensor * cur_gate_inp = llm_build_lora_mm(lctx, ctx0, model.layers[il].ffn_gate_inp_shexp, cur);
+                cb(cur_gate_inp, "ffn_shexp_gate_inp", il);
 
-    cur = ggml_mul(ctx, cur, g);
-    cur = llm_build_lora_mm(lctx, ctx, layer->time_mix_output, cur);
+                // sigmoid
+                ggml_tensor * cur_gate = ggml_div(ctx0, ggml_silu(ctx0, cur_gate_inp), cur_gate_inp);
+                cb(cur_gate, "ffn_shexp_gate", il);
 
-    return ggml_reshape_3d(ctx, cur, n_embd, n_seq_tokens, n_seqs);
-}
+                ggml_tensor * cur_ffn = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up_shexp,   NULL, NULL,
+                        model.layers[il].ffn_gate_shexp, NULL, NULL,
+                        model.layers[il].ffn_down_shexp, NULL, NULL,
+                        NULL,
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                cb(cur_ffn, "ffn_shexp", il);
 
-static struct ggml_tensor * llm_build_rwkv6_channel_mix(
-        struct llama_context & lctx,
-        struct ggml_context * ctx,
-        const struct llama_layer * layer,
-        struct ggml_tensor * cur,
-        struct ggml_tensor * x_prev) {
-    struct ggml_tensor * sx = ggml_sub(ctx, x_prev, cur);
-    struct ggml_tensor * xk = ggml_add(ctx, ggml_mul(ctx, sx, layer->channel_mix_lerp_k), cur);
-    struct ggml_tensor * xr = ggml_add(ctx, ggml_mul(ctx, sx, layer->channel_mix_lerp_r), cur);
+                ggml_tensor * ffn_shexp_out = ggml_mul(ctx0, cur_ffn, cur_gate);
+                cb(ffn_shexp_out, "ffn_shexp_out", il);
 
-    struct ggml_tensor * r = ggml_sigmoid(ctx, llm_build_lora_mm(lctx, ctx, layer->channel_mix_receptance, xr));
-    struct ggml_tensor * k = ggml_sqr(
-        ctx,
-        ggml_relu(
-            ctx,
-            llm_build_lora_mm(lctx, ctx, layer->channel_mix_key, xk)
-        )
-    );
+                moe_out = ggml_add(ctx0, moe_out, ffn_shexp_out);
+                cb(moe_out, "ffn_out", il);
 
-    return ggml_mul(ctx, r, llm_build_lora_mm(lctx, ctx, layer->channel_mix_value, k));
-}
+                cur = moe_out;
+            }
 
-struct llm_build_context {
-    const llama_model    & model;
-          llama_context  & lctx;
-    const llama_hparams  & hparams;
-    const llama_cparams  & cparams;
-    const llama_ubatch   & ubatch;
-    const llama_kv_cache & kv_self;
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-    const int64_t n_embd;
-    const int64_t n_layer;
-    const int64_t n_rot;
-    const int64_t n_ctx;       // user-specified context size (can be different from n_ctx_train)
-    const int64_t n_head;
-    const int64_t n_head_kv;
-    const int64_t n_embd_head_k;
-    const int64_t n_embd_k_gqa;
-    const int64_t n_embd_head_v;
-    const int64_t n_embd_v_gqa;
-    const int64_t n_expert;
-    const int64_t n_expert_used;
+            // input for next layer
+            inpL = cur;
+        }
 
-    const float freq_base;
-    const float freq_scale;
-    const float ext_factor;
-    const float attn_factor;
-    const float beta_fast;
-    const float beta_slow;
-    const float norm_eps;
-    const float norm_rms_eps;
+        cur = inpL;
 
-    const int32_t n_tokens;
-    const int32_t n_kv;     // size of KV cache to consider (n_kv <= kv_self.size)
-    const int32_t n_outputs;
-    const int32_t n_outputs_enc;
-    const int32_t kv_head;  // index of where we store new KV data in the cache
-    const int32_t n_ctx_orig;
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
+        cb(cur, "result_norm", -1);
 
-    const bool flash_attn;
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-    const enum llama_pooling_type pooling_type;
-    const enum llama_rope_type    rope_type;
+        ggml_build_forward_expand(gf, cur);
 
-    const llm_build_cb & cb;
+        return gf;
+    }
 
-    std::vector & buf_compute_meta;
+    struct ggml_cgraph * build_phi2() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-    struct ggml_context * ctx0 = nullptr;
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-    // TODO: consider making the entire interface noexcept
-    llm_build_context(
-        llama_context  & lctx,
-    const llama_ubatch & ubatch,
-    const llm_build_cb & cb,
-                  bool   worst_case) :
-        model            (lctx.model),
-        lctx             (lctx),
-        hparams          (model.hparams),
-        cparams          (lctx.cparams),
-        ubatch           (ubatch),
-        kv_self          (lctx.kv_self),
-        n_embd           (hparams.n_embd),
-        n_layer          (hparams.n_layer),
-        n_rot            (hparams.n_rot),
-        n_ctx            (cparams.n_ctx),
-        n_head           (hparams.n_head()),
-        n_head_kv        (hparams.n_head_kv()),
-        n_embd_head_k    (hparams.n_embd_head_k),
-        n_embd_k_gqa     (hparams.n_embd_k_gqa()),
-        n_embd_head_v    (hparams.n_embd_head_v),
-        n_embd_v_gqa     (hparams.n_embd_v_gqa()),
-        n_expert         (hparams.n_expert),
-        n_expert_used    (hparams.n_expert_used),
-        freq_base        (cparams.rope_freq_base),
-        freq_scale       (cparams.rope_freq_scale),
-        ext_factor       (cparams.yarn_ext_factor),
-        attn_factor      (cparams.yarn_attn_factor),
-        beta_fast        (cparams.yarn_beta_fast),
-        beta_slow        (cparams.yarn_beta_slow),
-        norm_eps         (hparams.f_norm_eps),
-        norm_rms_eps     (hparams.f_norm_rms_eps),
-        n_tokens         (ubatch.n_tokens),
-        n_kv             (worst_case ? kv_self.size : kv_self.n),
-        n_outputs        (worst_case ? n_tokens : lctx.n_outputs),
-        n_outputs_enc    (worst_case ? n_tokens : lctx.embd_enc.size() / hparams.n_embd),
-        kv_head          (worst_case ? (kv_self.recurrent ? 0 : kv_self.size - n_tokens) : kv_self.head),
-        n_ctx_orig       (cparams.n_ctx_orig_yarn),
-        flash_attn       (cparams.flash_attn),
-        pooling_type     (cparams.pooling_type),
-        rope_type        (hparams.rope_type),
-        cb               (cb),
-        buf_compute_meta (lctx.buf_compute_meta) {
-            // all initializations should be done in init()
-        }
-
-    void init() {
-        struct ggml_init_params params = {
-            /*.mem_size   =*/ buf_compute_meta.size(),
-            /*.mem_buffer =*/ buf_compute_meta.data(),
-            /*.no_alloc   =*/ true,
-        };
+        struct ggml_tensor * cur;
+        struct ggml_tensor * attn_norm_output;
+        struct ggml_tensor * ffn_output;
+        struct ggml_tensor * inpL;
 
-        ctx0 = ggml_init(params);
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-        lctx.inp_tokens      = nullptr;
-        lctx.inp_embd        = nullptr;
-        lctx.inp_pos         = nullptr;
-        lctx.inp_out_ids     = nullptr;
-        lctx.inp_KQ_mask     = nullptr;
-        lctx.inp_KQ_mask_swa = nullptr;
-        lctx.inp_K_shift     = nullptr;
-        lctx.inp_mean        = nullptr;
-        lctx.inp_cls         = nullptr;
-        lctx.inp_s_copy      = nullptr;
-        lctx.inp_s_mask      = nullptr;
-        lctx.inp_s_seq       = nullptr;
-        lctx.inp_pos_bucket    = nullptr;
-        lctx.inp_embd_enc      = nullptr;
-        lctx.inp_KQ_mask_cross = nullptr;
-    }
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-    void free() {
-        ggml_free(ctx0);
-        ctx0 = nullptr;
-    }
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-    struct ggml_cgraph * build_k_shift() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
+        for (int il = 0; il < n_layer; ++il) {
+            attn_norm_output = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm,
+                    model.layers[il].attn_norm_b,
+                    LLM_NORM, cb, il);
+            cb(attn_norm_output, "attn_norm", il);
 
-        GGML_ASSERT(kv_self.size == n_ctx);
+            // self-attention
+            {
+                struct ggml_tensor * Qcur = nullptr;
+                struct ggml_tensor * Kcur = nullptr;
+                struct ggml_tensor * Vcur = nullptr;
 
-        lctx.inp_K_shift = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_ctx);
-        cb(lctx.inp_K_shift, "K_shift", -1);
-        ggml_set_input(lctx.inp_K_shift);
+                if (model.layers[il].wqkv) {
+                    cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, attn_norm_output);
+                    cb(cur, "wqkv", il);
 
-        for (int il = 0; il < n_layer; ++il) {
-            const int64_t n_head_kv = hparams.n_head_kv(il);
-            const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(il);
-            struct ggml_tensor * rope_factors = build_rope_factors(il);
-            struct ggml_tensor * k =
-                ggml_view_3d(ctx0, kv_self.k_l[il],
-                    n_embd_head_k, n_head_kv, n_ctx,
-                    ggml_row_size(kv_self.k_l[il]->type, n_embd_head_k),
-                    ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa),
-                    0);
+                    cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
+                    cb(cur, "bqkv", il);
 
-            struct ggml_tensor * tmp;
-            if (ggml_is_quantized(k->type)) {
-                // dequantize to f32 -> RoPE -> quantize back
-                tmp = ggml_cast(ctx0, k, GGML_TYPE_F32);
-                cb(tmp, "K_f32", il);
-                for (auto & backend : lctx.backends) {
-                    // Figure out which backend KV cache belongs to
-                    if (ggml_backend_supports_buft(backend.get(), ggml_backend_buffer_get_type(kv_self.k_l[il]->buffer))) {
-                        ggml_backend_sched_set_tensor_backend(lctx.sched.get(), tmp, backend.get());
-                        break;
-                    }
+                    Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
+                    Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
+                    Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
+                } else {
+                    Qcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, attn_norm_output), model.layers[il].bq);
+                    Kcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, attn_norm_output), model.layers[il].bk);
+                    Vcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, attn_norm_output), model.layers[il].bv);
                 }
-                tmp = ggml_rope_ext_inplace(ctx0, tmp,
-                        lctx.inp_K_shift, rope_factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
-                cb(tmp, "K_shifted_f32", il);
-                tmp = ggml_cpy(ctx0, tmp, k);
-            } else {
-                // we rotate only the first n_rot dimensions
-                tmp = ggml_rope_ext_inplace(ctx0, k,
-                        lctx.inp_K_shift, rope_factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
-            }
-            cb(tmp, "K_shifted", il);
-            ggml_build_forward_expand(gf, tmp);
-        }
 
-        return gf;
-    }
+                cb(Qcur, "Qcur", il);
+                cb(Kcur, "Kcur", il);
+                cb(Vcur, "Vcur", il);
 
-    struct ggml_cgraph * build_defrag(const std::vector & ids) {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
 
-        for (uint32_t i = 0; i < ids.size(); ++i) {
-            const uint32_t id = ids[i];
+                Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig,
+                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-            if (i == id || id == ids.size()) {
-                continue;
-            }
+                // with phi2, we scale the Q to avoid precision issues
+                // ref: https://github.com/ml-explore/mlx-examples/blob/08e862336ade809bc37d1035f94b359e7d1a5152/phi2/phi2.py#L64-L66
+                Qcur = ggml_scale(ctx0, Qcur, 1.0f/sqrtf(float(n_embd_head)));
+                cb(Qcur, "Qcur", il);
 
-            uint32_t nm = 1;
+                Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig,
+                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
-            while (i + nm < ids.size() && ids[i + nm] == id + nm) {
-                nm++;
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, model.layers[il].bo,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f, cb, il);
             }
 
-            for (int il = 0; il < n_layer; ++il) {
-                const int64_t n_embd_k_gqa = hparams.n_embd_k_gqa(il);
-                const int64_t n_embd_v_gqa = hparams.n_embd_v_gqa(il);
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur              = ggml_get_rows(ctx0,              cur, inp_out_ids);
+                inpL             = ggml_get_rows(ctx0,             inpL, inp_out_ids);
+                attn_norm_output = ggml_get_rows(ctx0, attn_norm_output, inp_out_ids);
+            }
 
-                ggml_tensor * view_k_src = ggml_view_2d(ctx0, kv_self.k_l[il],
-                        n_embd_k_gqa, nm,
-                        ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa),
-                        ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa*i));
+            // FF
+            {
+                ffn_output = llm_build_ffn(ctx0, lctx, attn_norm_output,
+                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                        NULL,                      NULL,                        NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        NULL,
+                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
+                cb(ffn_output, "ffn_out", il);
+            }
 
-                ggml_tensor * view_k_dst = ggml_view_2d(ctx0, kv_self.k_l[il],
-                        n_embd_k_gqa, nm,
-                        ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa),
-                        ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa*id));
+            cur = ggml_add(ctx0, cur, ffn_output);
+            cur = ggml_add(ctx0, cur, inpL);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-                ggml_tensor * view_v_src;
-                ggml_tensor * view_v_dst;
+            // input for next layer
+            inpL = cur;
+        }
 
-                if (flash_attn) {
-                    // NOTE: the V cache is not transposed when using flash attention
-                    view_v_src = ggml_view_2d(ctx0, kv_self.v_l[il],
-                            n_embd_v_gqa, nm,
-                            ggml_row_size(kv_self.v_l[il]->type, n_embd_v_gqa),
-                            ggml_row_size(kv_self.v_l[il]->type, n_embd_v_gqa*i));
+        cur = llm_build_norm(ctx0, inpL, hparams,
+                model.output_norm,
+                model.output_norm_b,
+                LLM_NORM, cb, -1);
+        cb(cur, "result_norm", -1);
 
-                    view_v_dst = ggml_view_2d(ctx0, kv_self.v_l[il],
-                            n_embd_v_gqa, nm,
-                            ggml_row_size(kv_self.v_l[il]->type, n_embd_v_gqa),
-                            ggml_row_size(kv_self.v_l[il]->type, n_embd_v_gqa*id));
-                } else {
-                    view_v_src = ggml_view_2d(ctx0, kv_self.v_l[il],
-                            nm, n_embd_v_gqa,
-                            ggml_row_size(kv_self.v_l[il]->type, kv_self.size),
-                            ggml_row_size(kv_self.v_l[il]->type, i));
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output_no_bias", -1);
 
-                    view_v_dst = ggml_view_2d(ctx0, kv_self.v_l[il],
-                            nm, n_embd_v_gqa,
-                            ggml_row_size(kv_self.v_l[il]->type, kv_self.size),
-                            ggml_row_size(kv_self.v_l[il]->type, id));
-                }
+        cur = ggml_add(ctx0, cur, model.output_b);
+        cb(cur, "result_output", -1);
+        ggml_build_forward_expand(gf, cur);
+        return gf;
+    }
 
-                ggml_build_forward_expand(gf, ggml_cpy(ctx0, view_k_src, view_k_dst));
-                ggml_build_forward_expand(gf, ggml_cpy(ctx0, view_v_src, view_v_dst));
-            }
+    struct ggml_cgraph * build_phi3() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-            i += nm - 1;
-        }
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
-        //LLAMA_LOG_INFO("gf->n_nodes = %d\n", gf->n_nodes);
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
 
-        return gf;
-    }
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-    struct ggml_tensor * build_inp_pos() {
-        lctx.inp_pos = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_tokens);
-        cb(lctx.inp_pos, "inp_pos", -1);
-        ggml_set_input(lctx.inp_pos);
-        return lctx.inp_pos;
-    }
-
-    struct ggml_tensor * build_rope_factors(int il) {
-        // choose long/short freq factors based on the context size
-        const auto n_ctx_pre_seq = cparams.n_ctx / cparams.n_seq_max;
-
-        if (model.layers[il].rope_freqs != nullptr) {
-            return model.layers[il].rope_freqs;
-        }
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
-        if (n_ctx_pre_seq > hparams.n_ctx_orig_yarn) {
-            return model.layers[il].rope_long;
+        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
+        struct ggml_tensor * KQ_mask = nullptr;
+        if (hparams.n_swa == 0) {
+            // Phi-4 doesn't use sliding window attention
+            KQ_mask = build_inp_KQ_mask();
+        } else {
+            KQ_mask = build_inp_KQ_mask_swa();
         }
 
-        return model.layers[il].rope_short;
-    }
+        for (int il = 0; il < n_layer; ++il) {
+            auto residual = inpL;
 
-    struct ggml_tensor * build_inp_out_ids() {
-        lctx.inp_out_ids = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_outputs);
-        cb(lctx.inp_out_ids, "inp_out_ids", -1);
-        ggml_set_input(lctx.inp_out_ids);
-        return lctx.inp_out_ids;
-    }
+            // self-attention
+            {
+                // rope freq factors for 128k context
+                struct ggml_tensor * rope_factors = build_rope_factors(il);
 
-    struct ggml_tensor * build_inp_KQ_mask(bool causal = true) {
-        lctx.inp_KQ_mask = causal
-            ? ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv,     GGML_PAD(n_tokens, GGML_KQ_MASK_PAD))
-            : ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_tokens, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
-        cb(lctx.inp_KQ_mask, "KQ_mask", -1);
-        ggml_set_input(lctx.inp_KQ_mask);
+                struct ggml_tensor* attn_norm_output = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm,
+                    NULL,
+                    LLM_NORM_RMS, cb, il);
+                cb(attn_norm_output, "attn_norm", il);
 
-        return flash_attn ? ggml_cast(ctx0, lctx.inp_KQ_mask, GGML_TYPE_F16) : lctx.inp_KQ_mask;
-    }
+                struct ggml_tensor * Qcur = nullptr;
+                struct ggml_tensor * Kcur = nullptr;
+                struct ggml_tensor * Vcur = nullptr;
 
-    struct ggml_tensor * build_inp_KQ_mask_swa(bool causal = true) {
-        GGML_ASSERT(hparams.n_swa > 0);
+                if (model.layers[il].wqkv) {
+                    cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, attn_norm_output);
+                    cb(cur, "wqkv", il);
 
-        lctx.inp_KQ_mask_swa = causal
-            ? ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv,     GGML_PAD(n_tokens, GGML_KQ_MASK_PAD))
-            : ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_tokens, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
-        cb(lctx.inp_KQ_mask_swa, "KQ_mask_swa", -1);
-        ggml_set_input(lctx.inp_KQ_mask_swa);
+                    Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0 * sizeof(float) * (n_embd)));
+                    Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1 * sizeof(float) * (n_embd)));
+                    Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1 * sizeof(float) * (n_embd + n_embd_gqa)));
+                }
+                else {
+                    Qcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, attn_norm_output), model.layers[il].bq);
+                    Kcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, attn_norm_output), model.layers[il].bk);
+                    Vcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, attn_norm_output), model.layers[il].bv);
+                }
 
-        return flash_attn ? ggml_cast(ctx0, lctx.inp_KQ_mask_swa, GGML_TYPE_F16) : lctx.inp_KQ_mask_swa;
-    }
+                cb(Qcur, "Qcur", il);
+                cb(Kcur, "Kcur", il);
+                cb(Vcur, "Vcur", il);
 
-    struct ggml_tensor * build_inp_mean() {
-        lctx.inp_mean = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_tokens, n_tokens);
-        cb(lctx.inp_mean, "inp_mean", -1);
-        ggml_set_input(lctx.inp_mean);
-        return lctx.inp_mean;
-    }
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
 
-    struct ggml_tensor * build_inp_cls() {
-        lctx.inp_cls = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_tokens);
-        cb(lctx.inp_cls, "inp_cls", -1);
-        ggml_set_input(lctx.inp_cls);
-        return lctx.inp_cls;
-    }
+                Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, rope_factors, n_rot, rope_type, n_ctx_orig,
+                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
 
-    struct ggml_tensor * build_inp_s_copy() {
-        lctx.inp_s_copy = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_kv);
-        cb(lctx.inp_s_copy, "inp_s_copy", -1);
-        ggml_set_input(lctx.inp_s_copy);
-        return lctx.inp_s_copy;
-    }
+                Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd_head)));
+                cb(Qcur, "Qcur", il);
 
-    struct ggml_tensor * build_inp_s_mask() {
-        lctx.inp_s_mask = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, 1, n_kv);
-        cb(lctx.inp_s_mask, "inp_s_mask", -1);
-        ggml_set_input(lctx.inp_s_mask);
-        return lctx.inp_s_mask;
-    }
+                Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, rope_factors, n_rot, rope_type, n_ctx_orig,
+                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
-    struct ggml_cgraph * append_pooling(struct ggml_cgraph * gf) {
-        // find result_norm tensor for input
-        struct ggml_tensor * inp = nullptr;
-        for (int i = ggml_graph_n_nodes(gf) - 1; i >= 0; --i) {
-            inp = ggml_graph_node(gf, i);
-            if (strcmp(inp->name, "result_norm") == 0 || strcmp(inp->name, "result_embd") == 0) {
-                break;
-            } else {
-                inp = nullptr;
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, model.layers[il].bo,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f, cb, il);
             }
-        }
-        GGML_ASSERT(inp != nullptr && "missing result_norm/result_embd tensor");
 
-        struct ggml_tensor * cur;
+            if (il == n_layer - 1) {
+                // skip computing output for unused tokens
+                struct ggml_tensor* inp_out_ids = build_inp_out_ids();
+                cur = ggml_get_rows(ctx0, cur, inp_out_ids);
+                residual = ggml_get_rows(ctx0, residual, inp_out_ids);
+            }
 
-        switch (pooling_type) {
-            case LLAMA_POOLING_TYPE_NONE:
-                {
-                    cur = inp;
-                } break;
-            case LLAMA_POOLING_TYPE_MEAN:
-                {
-                    struct ggml_tensor * inp_mean = build_inp_mean();
-                    cur = ggml_mul_mat(ctx0, ggml_cont(ctx0, ggml_transpose(ctx0, inp)), inp_mean);
-                } break;
-            case LLAMA_POOLING_TYPE_CLS:
-            case LLAMA_POOLING_TYPE_LAST:
-                {
-                    struct ggml_tensor * inp_cls = build_inp_cls();
-                    cur = ggml_get_rows(ctx0, inp, inp_cls);
-                } break;
-            case LLAMA_POOLING_TYPE_RANK:
-                {
-                    struct ggml_tensor * inp_cls = build_inp_cls();
-                    inp = ggml_get_rows(ctx0, inp, inp_cls);
+            cur = ggml_add(ctx0, cur, residual);
+            residual = cur;
 
-                    // classification head
-                    // https://github.com/huggingface/transformers/blob/5af7d41e49bbfc8319f462eb45253dcb3863dfb7/src/transformers/models/roberta/modeling_roberta.py#L1566
-                    GGML_ASSERT(model.cls       != nullptr);
-                    GGML_ASSERT(model.cls_b     != nullptr);
+            cur = llm_build_norm(ctx0, cur, hparams,
+                model.layers[il].ffn_norm, NULL,
+                LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_norm", il);
 
-                    cur = ggml_add (ctx0, ggml_mul_mat(ctx0, model.cls, inp), model.cls_b);
-                    cur = ggml_tanh(ctx0, cur);
+            // FF
+            // special-case: the up and gate tensors are merged into a single tensor
+            // TOOD: support into llm_build_ffn
+            {
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        NULL,                      NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
+                        NULL,
+                        LLM_FFN_SWIGLU, LLM_FFN_SEQ, cb, il);
+                cb(cur, "ffn_out", il);
+            }
 
-                    // some models don't have `cls_out`, for example: https://huggingface.co/jinaai/jina-reranker-v1-tiny-en
-                    // https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/blob/cb5347e43979c3084a890e3f99491952603ae1b7/modeling_bert.py#L884-L896
-                    if (model.cls_out) {
-                        GGML_ASSERT(model.cls_out_b != nullptr);
+            cur = ggml_add(ctx0, residual, cur);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
-                        cur = ggml_add (ctx0, ggml_mul_mat(ctx0, model.cls_out, cur), model.cls_out_b);
-                    }
-                } break;
-            default:
-                {
-                    GGML_ABORT("unknown pooling type");
-                }
+            // input for next layer
+            inpL = cur;
         }
 
-        cb(cur, "result_embd_pooled", -1);
+        cur = llm_build_norm(ctx0, inpL, hparams,
+            model.output_norm,
+            NULL,
+            LLM_NORM_RMS, cb, -1);
+        cb(cur, "result_norm", -1);
+
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
 
         return gf;
     }
 
-    struct ggml_tensor * llm_build_pos_bucket(bool causal) {
-        if (causal) {
-            lctx.inp_pos_bucket = ggml_new_tensor_2d(ctx0, GGML_TYPE_I32, n_kv,     n_tokens);
-        } else {
-            lctx.inp_pos_bucket = ggml_new_tensor_2d(ctx0, GGML_TYPE_I32, n_tokens, n_tokens);
-        }
-
-        ggml_set_input(lctx.inp_pos_bucket);
-        cb(lctx.inp_pos_bucket, "pos_bucket", -1);
-
-        return lctx.inp_pos_bucket;
-    }
-
-    struct ggml_tensor * llm_build_pos_bias(struct ggml_tensor * pos_bucket, struct ggml_tensor * attn_rel_b) {
-        struct ggml_tensor * pos_bucket_1d = ggml_view_1d(ctx0, pos_bucket, pos_bucket->ne[0] * pos_bucket->ne[1], 0);
-        cb(pos_bucket_1d, "pos_bucket_1d", -1);
-
-        struct ggml_tensor * pos_bias = ggml_get_rows(ctx0, attn_rel_b, pos_bucket_1d);
-        cb(pos_bias, "pos_bias", -1);
-
-        pos_bias = ggml_view_3d(ctx0, pos_bias, pos_bias->ne[0], lctx.inp_pos_bucket->ne[0], lctx.inp_pos_bucket->ne[1], ggml_element_size(pos_bias) * pos_bias->ne[0], ggml_element_size(pos_bias) * pos_bias->ne[0] * lctx.inp_pos_bucket->ne[0],  0);
-        cb(pos_bias, "pos_bias", -1);
-
-        pos_bias = ggml_permute(ctx0, pos_bias, 2, 0, 1, 3);
-        cb(pos_bias, "pos_bias", -1);
-
-        pos_bias = ggml_cont(ctx0, pos_bias);
-        cb(pos_bias, "pos_bias", -1);
-
-        return pos_bias;
-    }
-
-    struct ggml_tensor * llm_build_inp_embd_enc() {
-        const int64_t n_embd = hparams.n_embd;
-        lctx.inp_embd_enc = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_embd, n_outputs_enc);
-        ggml_set_input(lctx.inp_embd_enc);
-        cb(lctx.inp_embd_enc, "embd_enc", -1);
-        return lctx.inp_embd_enc;
-    }
-
-    struct ggml_tensor * llm_build_inp_KQ_mask_cross() {
-        lctx.inp_KQ_mask_cross = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_outputs_enc, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
-        ggml_set_input(lctx.inp_KQ_mask_cross);
-        cb(lctx.inp_KQ_mask_cross, "KQ_mask_cross", -1);
-        return lctx.inp_KQ_mask_cross;
-    }
-
-    struct ggml_cgraph * build_llama() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        // mutable variable, needed during the last layer of the computation to skip unused tokens
-        int32_t n_tokens = this->n_tokens;
+    struct ggml_cgraph * build_plamo() {
+        struct ggml_cgraph * gf = ggml_new_graph(ctx0);
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@@ -10932,9 +6377,7 @@ struct llm_build_context {
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-        const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
         for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
 
             // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
@@ -10942,109 +6385,61 @@ struct llm_build_context {
                     LLM_NORM_RMS, cb, il);
             cb(cur, "attn_norm", il);
 
+            struct ggml_tensor * attention_norm = cur;
+
             // self-attention
             {
-                // rope freq factors for llama3; may return nullptr for llama2 and other models
-                struct ggml_tensor * rope_factors = build_rope_factors(il);
-
                 // compute Q and K and RoPE them
                 struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
                 cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
 
                 struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
                 cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
 
                 struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
                 cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
 
                 Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, rope_factors,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
+                        ctx0, ggml_reshape_3d(ctx0, Qcur, n_rot, n_head,    n_tokens), inp_pos, nullptr,
+                        n_embd_head, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow);
                 cb(Qcur, "Qcur", il);
 
                 Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, rope_factors,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
+                        ctx0, ggml_reshape_3d(ctx0, Kcur, n_rot, n_head_kv, n_tokens), inp_pos, nullptr,
+                        n_embd_head, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow);
                 cb(Kcur, "Kcur", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, kq_scale, cb, il);
+                        model.layers[il].wo, NULL,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
             }
+            struct ggml_tensor * sa_out = cur;
+
+            cur = attention_norm;
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                n_tokens = n_outputs;
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-
-            // For Granite architecture
-            if (hparams.f_residual_scale) {
-                cur = ggml_scale(ctx0, cur, hparams.f_residual_scale);
+                cur    = ggml_get_rows(ctx0,    cur, inp_out_ids);
+                sa_out = ggml_get_rows(ctx0, sa_out, inp_out_ids);
+                inpL   = ggml_get_rows(ctx0,   inpL, inp_out_ids);
             }
 
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
             // feed-forward network
-            if (model.layers[il].ffn_gate_inp == nullptr) {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "ffn_norm", il);
-
+            {
                 cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        model.layers[il].ffn_gate, NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
                         NULL,
                         LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
                 cb(cur, "ffn_out", il);
-            } else {
-                // MoE branch
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = llm_build_moe_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_gate_inp,
-                        model.layers[il].ffn_up_exps,
-                        model.layers[il].ffn_gate_exps,
-                        model.layers[il].ffn_down_exps,
-                        n_expert, n_expert_used,
-                        LLM_FFN_SILU, true,
-                        false, 0.0,
-                        cb, il);
-                cb(cur, "ffn_moe_out", il);
-            }
-
-            // For Granite architecture
-            if (hparams.f_residual_scale) {
-                cur = ggml_scale(ctx0, cur, hparams.f_residual_scale);
             }
 
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
+            cur = ggml_add(ctx0, cur, sa_out);
+            cur = ggml_add(ctx0, cur, inpL);
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
@@ -11061,12 +6456,6 @@ struct llm_build_context {
 
         // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-
-        // For Granite architecture
-        if (hparams.f_logit_scale) {
-            cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_logit_scale);
-        }
-
         cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
@@ -11074,94 +6463,86 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_baichuan() {
+    struct ggml_cgraph * build_gpt2() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
         struct ggml_tensor * cur;
+        struct ggml_tensor * pos;
         struct ggml_tensor * inpL;
 
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
         // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = model.type == MODEL_7B ? build_inp_pos() : nullptr;
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
+        pos = ggml_get_rows(ctx0, model.pos_embd, inp_pos);
+        cb(pos, "pos_embd", -1);
+
+        inpL = ggml_add(ctx0, inpL, pos);
+        cb(inpL, "inpL", -1);
 
+        for (int il = 0; il < n_layer; ++il) {
             cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
+                    model.layers[il].attn_norm,
+                    model.layers[il].attn_norm_b,
+                    LLM_NORM, cb, il);
             cb(cur, "attn_norm", il);
 
             // self-attention
             {
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
+                cb(cur, "bqkv", il);
 
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
+                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
 
-                switch (model.type) {
-                    case MODEL_7B:
-                        Qcur = ggml_rope_ext(
-                            ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
-                            n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                            ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-                        Kcur = ggml_rope_ext(
-                            ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
-                            n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                            ext_factor, attn_factor, beta_fast, beta_slow
-                        );
-                        break;
-                    case MODEL_13B:
-                        Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd/n_head, n_head, n_tokens);
-                        Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd/n_head, n_head, n_tokens);
-                        break;
-                    default:
-                        GGML_ABORT("fatal error");
-                }
                 cb(Qcur, "Qcur", il);
                 cb(Kcur, "Kcur", il);
+                cb(Vcur, "Vcur", il);
+
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
+                        model.layers[il].wo, model.layers[il].bo,
                         Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
             }
 
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            // add the input
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
             cb(ffn_inp, "ffn_inp", il);
 
-            // feed-forward network
+            // FF
             {
                 cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
+                        model.layers[il].ffn_norm,
+                        model.layers[il].ffn_norm_b,
+                        LLM_NORM, cb, il);
                 cb(cur, "ffn_norm", il);
 
                 cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
+                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                        NULL,                      NULL,                        NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
                         NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
                 cb(cur, "ffn_out", il);
             }
 
@@ -11173,14 +6554,12 @@ struct llm_build_context {
             inpL = cur;
         }
 
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
+        cur = llm_build_norm(ctx0, inpL, hparams,
+                model.output_norm,
+                model.output_norm_b,
+                LLM_NORM, cb, -1);
         cb(cur, "result_norm", -1);
 
-        // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
         cb(cur, "result_output", -1);
 
@@ -11189,10 +6568,11 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_xverse() {
+    struct ggml_cgraph * build_codeshell() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
         GGML_ASSERT(n_embd_head == hparams.n_rot);
 
@@ -11208,65 +6588,72 @@ struct llm_build_context {
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
         for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
             cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
+                    model.layers[il].attn_norm,
+                    model.layers[il].attn_norm_b,
+                    LLM_NORM, cb, il);
             cb(cur, "attn_norm", il);
 
             // self-attention
             {
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
+                cb(cur, "bqkv", il);
 
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                struct ggml_tensor * tmpq = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
+                struct ggml_tensor * tmpk = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
+
+                cb(tmpq, "tmpq", il);
+                cb(tmpk, "tmpk", il);
                 cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                struct ggml_tensor * Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, tmpq, n_embd_head, n_head,    n_tokens), inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
                 cb(Qcur, "Qcur", il);
 
-                Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                struct ggml_tensor * Kcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, tmpk, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
                 cb(Kcur, "Kcur", il);
+
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
+                        model.layers[il].wo, model.layers[il].bo,
                         Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur   = ggml_get_rows(ctx0,      cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
             }
 
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            // add the input
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
             cb(ffn_inp, "ffn_inp", il);
 
-            // feed-forward network
+            // FF
             {
                 cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
+                        model.layers[il].ffn_norm,
+                        model.layers[il].ffn_norm_b,
+                        LLM_NORM, cb, il);
                 cb(cur, "ffn_norm", il);
 
                 cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
+                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                        NULL,                      NULL,                        NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
                         NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
                 cb(cur, "ffn_out", il);
             }
 
@@ -11278,12 +6665,12 @@ struct llm_build_context {
             inpL = cur;
         }
 
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams, model.output_norm, NULL, LLM_NORM_RMS, cb, -1);
+        cur = llm_build_norm(ctx0, inpL, hparams,
+                model.output_norm,
+                model.output_norm_b,
+                LLM_NORM, cb, -1);
         cb(cur, "result_norm", -1);
 
-        // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
         cb(cur, "result_output", -1);
 
@@ -11292,11 +6679,10 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_falcon() {
+    struct ggml_cgraph * build_orion() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
         GGML_ASSERT(n_embd_head == hparams.n_rot);
 
@@ -11312,51 +6698,49 @@ struct llm_build_context {
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
         for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * attn_norm;
+            struct ggml_tensor * inpSA = inpL;
 
-            attn_norm = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
+            // norm
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, model.layers[il].attn_norm_b,
                     LLM_NORM, cb, il);
-            cb(attn_norm, "attn_norm", il);
+            cb(cur, "attn_norm", il);
 
             // self-attention
             {
-                if (model.layers[il].attn_norm_2) {
-                    // Falcon-40B
-                    cur = llm_build_norm(ctx0, inpL, hparams,
-                            model.layers[il].attn_norm_2,
-                            model.layers[il].attn_norm_2_b,
-                            LLM_NORM, cb, il);
-                    cb(cur, "attn_norm_2", il);
-                } else {
-                    cur = attn_norm;
-                }
-
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
-
+                // compute Q and K and RoPE them
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
                 cb(Qcur, "Qcur", il);
+                // if (model.layers[il].bq) {
+                //     Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                //     cb(Qcur, "Qcur", il);
+                // }
+
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
                 cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+                // if (model.layers[il].bk) {
+                //     Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                //     cb(Kcur, "Kcur", il);
+                // }
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
+                // if (model.layers[il].bv) {
+                //     Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                //     cb(Vcur, "Vcur", il);
+                // }
 
-                // using mode = 2 for neox mode
                 Qcur = ggml_rope_ext(
-                    ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig,
-                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
                 );
                 cb(Qcur, "Qcur", il);
 
                 Kcur = ggml_rope_ext(
-                    ctx0, Kcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig,
-                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
                 );
                 cb(Kcur, "Kcur", il);
 
@@ -11368,26 +6752,28 @@ struct llm_build_context {
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur       = ggml_get_rows(ctx0,       cur, inp_out_ids);
-                inpL      = ggml_get_rows(ctx0,      inpL, inp_out_ids);
-                attn_norm = ggml_get_rows(ctx0, attn_norm, inp_out_ids);
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
             }
 
-            struct ggml_tensor * ffn_inp = cur;
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
-            // feed forward
-            {
-                cur = llm_build_ffn(ctx0, lctx, attn_norm, // !! use the attn norm, not the result
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        NULL,                      NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
-                cb(cur, "ffn_out", il);
-            }
+            // feed-forward network
+            cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                    model.layers[il].ffn_norm, model.layers[il].ffn_norm_b,
+                    LLM_NORM, cb, il);
+            cb(cur, "ffn_norm", il);
+
+            cur = llm_build_ffn(ctx0, lctx, cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+            cb(cur, "ffn_out", il);
 
             cur = ggml_add(ctx0, cur, ffn_inp);
-            cur = ggml_add(ctx0, cur, inpL);
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
@@ -11397,13 +6783,12 @@ struct llm_build_context {
 
         cur = inpL;
 
-        // norm
         cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm,
-                model.output_norm_b,
+                model.output_norm, model.output_norm_b,
                 LLM_NORM, cb, -1);
         cb(cur, "result_norm", -1);
 
+        // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
         cb(cur, "result_output", -1);
 
@@ -11412,12 +6797,9 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_grok() {
+    struct ggml_cgraph * build_internlm2() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        // mutable variable, needed during the last layer of the computation to skip unused tokens
-        int32_t n_tokens = this->n_tokens;
-
         const int64_t n_embd_head = hparams.n_embd_head_v;
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
         GGML_ASSERT(n_embd_head == hparams.n_rot);
@@ -11427,9 +6809,6 @@ struct llm_build_context {
 
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-        // multiply by embedding_multiplier_scale of 78.38367176906169
-        inpL = ggml_scale(ctx0, inpL, 78.38367176906169f);
-
         // inp_pos - contains the positions
         struct ggml_tensor * inp_pos = build_inp_pos();
 
@@ -11445,7 +6824,6 @@ struct llm_build_context {
                     LLM_NORM_RMS, cb, il);
             cb(cur, "attn_norm", il);
 
-
             // self-attention
             {
                 // compute Q and K and RoPE them
@@ -11471,7 +6849,7 @@ struct llm_build_context {
                 }
 
                 Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens), inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
@@ -11486,60 +6864,34 @@ struct llm_build_context {
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
                         model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f, cb, il);
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                n_tokens = n_outputs;
                 cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
                 inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
             }
 
-            // Grok
-            // if attn_out_norm is present then apply it before adding the input
-            if (model.layers[il].attn_out_norm) {
-                cur = llm_build_norm(ctx0, cur, hparams,
-                        model.layers[il].attn_out_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "attn_out_norm", il);
-            }
-
             struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
             cb(ffn_inp, "ffn_inp", il);
 
             // feed-forward network
-            // MoE branch
             cur = llm_build_norm(ctx0, ffn_inp, hparams,
                     model.layers[il].ffn_norm, NULL,
                     LLM_NORM_RMS, cb, il);
             cb(cur, "ffn_norm", il);
 
-            cur = llm_build_moe_ffn(ctx0, lctx, cur,
-                    model.layers[il].ffn_gate_inp,
-                    model.layers[il].ffn_up_exps,
-                    model.layers[il].ffn_gate_exps,
-                    model.layers[il].ffn_down_exps,
-                    n_expert, n_expert_used,
-                    LLM_FFN_GELU, true,
-                    false, 0.0,
-                    cb, il);
-            cb(cur, "ffn_moe_out", il);
-
-            // Grok
-            // if layer_out_norm is present then apply it before adding the input
-            // Idea: maybe ffn_out_norm is a better name
-            if (model.layers[il].layer_out_norm) {
-                cur = llm_build_norm(ctx0, cur, hparams,
-                        model.layers[il].layer_out_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "layer_out_norm", il);
-            }
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = llm_build_ffn(ctx0, lctx, cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
             cb(cur, "ffn_out", il);
 
+            cur = ggml_add(ctx0, cur, ffn_inp);
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
@@ -11556,12 +6908,6 @@ struct llm_build_context {
 
         // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-
-        // Grok
-        // multiply logits by output_multiplier_scale of 0.5773502691896257
-
-        cur = ggml_scale(ctx0, cur, 0.5773502691896257f);
-
         cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
@@ -11569,22 +6915,28 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_dbrx() {
+    struct ggml_cgraph * build_minicpm3() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        // mutable variable, needed during the last layer of the computation to skip unused tokens
-        int32_t n_tokens = this->n_tokens;
+        //TODO: if the model varies, these parameters need to be read from the model
+        const int64_t n_embd_base = 256;
+        const float scale_embd  = 12.0f;
+        const float scale_depth = 1.4f;
+        const float kq_scale = 1.0f / sqrtf(float(hparams.n_embd_head_k));
 
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+        const uint32_t n_embd_head_qk_rope = hparams.n_rot;
+        const uint32_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot;
+        const uint32_t kv_lora_rank = hparams.n_lora_kv;
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
 
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
+        // scale the input embeddings
+        inpL = ggml_scale(ctx0, inpL, scale_embd);
+        cb(inpL, "inp_scaled", -1);
+
         // inp_pos - contains the positions
         struct ggml_tensor * inp_pos = build_inp_pos();
 
@@ -11594,189 +6946,155 @@ struct llm_build_context {
         for (int il = 0; il < n_layer; ++il) {
             struct ggml_tensor * inpSA = inpL;
 
+            struct ggml_tensor * rope_factors = build_rope_factors(il);
             // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
-                                 model.layers[il].attn_norm, NULL,
-                                 LLM_NORM, cb, il);
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
             cb(cur, "attn_norm", il);
 
-            // self-attention
+            // self_attention
             {
-                struct ggml_tensor * Qcur = nullptr;
-                struct ggml_tensor * Kcur = nullptr;
-                struct ggml_tensor * Vcur = nullptr;
-
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
+                struct ggml_tensor * q = NULL;
+                // {n_embd, q_lora_rank} * {n_embd, n_tokens} -> {q_lora_rank, n_tokens}
+                q = ggml_mul_mat(ctx0, model.layers[il].wq_a, cur);
+                cb(q, "q", il);
 
-                cur = ggml_clamp(ctx0, cur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
-                cb(cur, "wqkv_clamped", il);
+                q = llm_build_norm(ctx0, q, hparams,
+                        model.layers[il].attn_q_a_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(q, "q", il);
 
-                Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
-                Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
-                Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
+                // {q_lora_rank, n_head * hparams.n_embd_head_k} * {q_lora_rank, n_tokens} -> {n_head * hparams.n_embd_head_k, n_tokens}
+                q = ggml_mul_mat(ctx0, model.layers[il].wq_b, q);
+                cb(q, "q", il);
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+                // split into {n_head * n_embd_head_qk_nope, n_tokens}
+                struct ggml_tensor * q_nope = ggml_view_3d(ctx0, q, n_embd_head_qk_nope, n_head, n_tokens,
+                        ggml_row_size(q->type, hparams.n_embd_head_k),
+                        ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
+                        0);
+                cb(q_nope, "q_nope", il);
 
-                Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur", il);
+                // and {n_head * n_embd_head_qk_rope, n_tokens}
+                struct ggml_tensor * q_pe = ggml_view_3d(ctx0, q, n_embd_head_qk_rope, n_head, n_tokens,
+                        ggml_row_size(q->type, hparams.n_embd_head_k),
+                        ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
+                        ggml_row_size(q->type, n_embd_head_qk_nope));
+                cb(q_pe, "q_pe", il);
 
-                Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur", il);
+                // {n_embd, kv_lora_rank + n_embd_head_qk_rope} * {n_embd, n_tokens} -> {kv_lora_rank + n_embd_head_qk_rope, n_tokens}
+                struct ggml_tensor * kv_pe_compresseed = ggml_mul_mat(ctx0, model.layers[il].wkv_a_mqa, cur);
+                cb(kv_pe_compresseed, "kv_pe_compresseed", il);
 
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-            }
+                // split into {kv_lora_rank, n_tokens}
+                struct ggml_tensor * kv_compressed = ggml_view_2d(ctx0, kv_pe_compresseed, kv_lora_rank, n_tokens,
+                        kv_pe_compresseed->nb[1],
+                        0);
+                cb(kv_compressed, "kv_compressed", il);
 
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                n_tokens = n_outputs;
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
+                // and {n_embd_head_qk_rope, n_tokens}
+                struct ggml_tensor * k_pe = ggml_view_3d(ctx0, kv_pe_compresseed, n_embd_head_qk_rope, 1, n_tokens,
+                        kv_pe_compresseed->nb[1],
+                        kv_pe_compresseed->nb[1],
+                        ggml_row_size(kv_pe_compresseed->type, kv_lora_rank));
+                cb(k_pe, "k_pe", il);
 
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            // MoE branch
-            cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                                 model.layers[il].attn_out_norm, NULL,
-                                 LLM_NORM, cb, il);
-            cb(cur, "attn_out_norm", il);
-
-            cur = llm_build_moe_ffn(ctx0, lctx, cur,
-                    model.layers[il].ffn_gate_inp,
-                    model.layers[il].ffn_up_exps,
-                    model.layers[il].ffn_gate_exps,
-                    model.layers[il].ffn_down_exps,
-                    n_expert, n_expert_used,
-                    LLM_FFN_SILU, true,
-                    false, 0.0,
-                    cb, il);
-            cb(cur, "ffn_moe_out", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                             model.output_norm, NULL,
-                             LLM_NORM, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        // lm_head
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    struct ggml_cgraph * build_starcoder() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
+                kv_compressed = ggml_cont(ctx0, kv_compressed); // TODO: the CUDA backend does not support non-contiguous norm
+                kv_compressed = llm_build_norm(ctx0, kv_compressed, hparams,
+                        model.layers[il].attn_kv_a_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(kv_compressed, "kv_compressed", il);
 
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
+                // {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)} * {kv_lora_rank, n_tokens} -> {n_head * (n_embd_head_qk_nope + n_embd_head_v), n_tokens}
+                struct ggml_tensor * kv = ggml_mul_mat(ctx0, model.layers[il].wkv_b, kv_compressed);
+                cb(kv, "kv", il);
 
-        struct ggml_tensor * pos = ggml_get_rows(ctx0, model.pos_embd, inp_pos);
-        cb(pos, "pos_embd", -1);
+                // split into {n_head * n_embd_head_qk_nope, n_tokens}
+                struct ggml_tensor * k_nope = ggml_view_3d(ctx0, kv, n_embd_head_qk_nope, n_head, n_tokens,
+                        ggml_row_size(kv->type, n_embd_head_qk_nope + hparams.n_embd_head_v),
+                        ggml_row_size(kv->type, n_head * (n_embd_head_qk_nope + hparams.n_embd_head_v)),
+                        0);
+                cb(k_nope, "k_nope", il);
 
-        inpL = ggml_add(ctx0, inpL, pos);
-        cb(inpL, "inpL", -1);
+                // and {n_head * n_embd_head_v, n_tokens}
+                struct ggml_tensor * v_states = ggml_view_3d(ctx0, kv, hparams.n_embd_head_v, n_head, n_tokens,
+                        ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)),
+                        ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)*n_head),
+                        ggml_row_size(kv->type, (n_embd_head_qk_nope)));
+                cb(v_states, "v_states", il);
 
-        for (int il = 0; il < n_layer; ++il) {
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, cb, il);
-            cb(cur, "attn_norm", il);
+                v_states = ggml_cont(ctx0, v_states);
+                cb(v_states, "v_states", il);
 
-            // self-attention
-            {
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
+                v_states = ggml_view_2d(ctx0, v_states, hparams.n_embd_head_v * n_head, n_tokens,
+                    ggml_row_size(kv->type, hparams.n_embd_head_v * n_head),
+                    0);
+                cb(v_states, "v_states", il);
 
-                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                cb(cur, "bqkv", il);
+                q_pe = ggml_cont(ctx0, q_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
+                q_pe = ggml_rope_ext(
+                    ctx0, q_pe, inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(q_pe, "q_pe", il);
 
-                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
+                // shared RoPE key
+                k_pe = ggml_cont(ctx0, k_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
+                k_pe = ggml_rope_ext(
+                    ctx0, k_pe, inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(k_pe, "k_pe", il);
 
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+                struct ggml_tensor * q_states = ggml_concat(ctx0, q_nope, q_pe, 0);
+                cb(q_states, "q_states", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
+                struct ggml_tensor * k_states = ggml_concat(ctx0, k_nope, ggml_repeat(ctx0, k_pe, q_pe), 0);
+                cb(k_states, "k_states", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+                        model.layers[il].wo, NULL,
+                        k_states, v_states, q_states, KQ_mask, n_tokens, kv_head, n_kv, kq_scale, cb, il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
             }
 
-            // add the input
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
+            // scale_res - scale the hidden states for residual connection
+            const float scale_res = scale_depth/sqrtf(float(n_layer));
+            cur = ggml_scale(ctx0, cur, scale_res);
+            cb(cur, "hidden_scaled", il);
+
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
             cb(ffn_inp, "ffn_inp", il);
 
-            // FF
+            // feed-forward network
             {
                 cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm,
-                        model.layers[il].ffn_norm_b,
-                        LLM_NORM, cb, il);
+                        model.layers[il].ffn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
                 cb(cur, "ffn_norm", il);
 
                 cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        NULL,                      NULL,                        NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        model.layers[il].ffn_gate, NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
                         NULL,
-                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
                 cb(cur, "ffn_out", il);
             }
 
+            // scale the hidden states for residual connection
+            cur = ggml_scale(ctx0, cur, scale_res);
+            cb(cur, "hidden_scaled_ffn", il);
+
             cur = ggml_add(ctx0, cur, ffn_inp);
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
@@ -11785,12 +7103,19 @@ struct llm_build_context {
             inpL = cur;
         }
 
-        cur = llm_build_norm(ctx0, inpL, hparams,
-                model.output_norm,
-                model.output_norm_b,
-                LLM_NORM, cb, -1);
+        cur = inpL;
+
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
         cb(cur, "result_norm", -1);
 
+        // lm_head scaling
+        const float scale_lmhead = float(n_embd_base)/float(n_embd);
+        cur = ggml_scale(ctx0, cur, scale_lmhead);
+        cb(cur, "lmhead_scaling", -1);
+
+        // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
         cb(cur, "result_output", -1);
 
@@ -11799,23 +7124,27 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_refact() {
+    struct ggml_cgraph * build_gemma() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        const int64_t n_embd_head_k = hparams.n_embd_head_k;
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
 
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
+        inpL = ggml_scale(ctx0, inpL, sqrtf(n_embd));
+        cb(inpL, "inp_scaled", -1);
+
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
+
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
         for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
+            // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
                     model.layers[il].attn_norm, NULL,
                     LLM_NORM_RMS, cb, il);
@@ -11823,6 +7152,7 @@ struct llm_build_context {
 
             // self-attention
             {
+                // compute Q and K and RoPE them
                 struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
                 cb(Qcur, "Qcur", il);
 
@@ -11832,44 +7162,53 @@ struct llm_build_context {
                 struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
                 cb(Vcur, "Vcur", il);
 
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                cb(Kcur, "Kcur", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+                Qcur = ggml_rope_ext(
+                        ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head_k, n_head,    n_tokens), inp_pos, nullptr,
+                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow);
                 cb(Qcur, "Qcur", il);
 
+                Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd_head_k)));
+                cb(Qcur, "Qcur_scaled", il);
+
+                Kcur = ggml_rope_ext(
+                        ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head_k, n_head_kv, n_tokens), inp_pos, nullptr,
+                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow);
+                cb(Kcur, "Kcur", il);
+
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
                         model.layers[il].wo, NULL,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f, cb, il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
             }
 
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            struct ggml_tensor * sa_out = ggml_add(ctx0, cur, inpL);
+            cb(sa_out, "sa_out", il);
+
+            cur = llm_build_norm(ctx0, sa_out, hparams,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_norm", il);
 
             // feed-forward network
             {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "ffn_norm", il);
-
                 cur = llm_build_ffn(ctx0, lctx, cur,
                         model.layers[il].ffn_up,   NULL, NULL,
                         model.layers[il].ffn_gate, NULL, NULL,
                         model.layers[il].ffn_down, NULL, NULL,
                         NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                        LLM_FFN_GELU, LLM_FFN_PAR, cb, il);
                 cb(cur, "ffn_out", il);
             }
 
-            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = ggml_add(ctx0, cur, sa_out);
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
@@ -11893,186 +7232,114 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_bert() {
+    struct ggml_cgraph * build_gemma2() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
-
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        const int64_t n_embd_head_k = hparams.n_embd_head_k;
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
-        struct ggml_tensor * inp_pos = nullptr;
-
-        if (model.arch != LLM_ARCH_JINA_BERT_V2) {
-            inp_pos = build_inp_pos();
-        }
 
-        // construct input embeddings (token, type, position)
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-        // token types are hardcoded to zero ("Sentence A")
-        struct ggml_tensor * type_row0 = ggml_view_1d(ctx0, model.type_embd, n_embd, 0);
-        inpL = ggml_add(ctx0, inpL, type_row0);
-        if (model.arch == LLM_ARCH_BERT) {
-            inpL = ggml_add(ctx0, ggml_get_rows(ctx0, model.pos_embd, inp_pos), inpL);
-        }
-        cb(inpL, "inp_embd", -1);
+        inpL = ggml_scale(ctx0, inpL, sqrtf(n_embd));
+        cb(inpL, "inp_scaled", -1);
 
-        // embed layer norm
-        inpL = llm_build_norm(ctx0, inpL, hparams, model.tok_norm, model.tok_norm_b, LLM_NORM, cb, -1);
-        cb(inpL, "inp_norm", -1);
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask(false);
+        // gemma 2 requires different mask for layers using sliding window (SWA)
+        struct ggml_tensor * KQ_mask     = build_inp_KQ_mask(true);
+        struct ggml_tensor * KQ_mask_swa = build_inp_KQ_mask_swa(true);
 
-        // iterate layers
         for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * cur = inpL;
+            // (il % 2) layers use SWA
+            struct ggml_tensor * KQ_mask_l = (il % 2 == 0) ? KQ_mask_swa : KQ_mask;
 
-            struct ggml_tensor * Qcur;
-            struct ggml_tensor * Kcur;
-            struct ggml_tensor * Vcur;
+            // norm
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm", il);
 
             // self-attention
-            if (model.arch == LLM_ARCH_BERT || model.arch == LLM_ARCH_JINA_BERT_V2) {
-                Qcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur), model.layers[il].bq);
+            {
+                // compute Q and K and RoPE them
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
                 cb(Qcur, "Qcur", il);
 
-                if (model.layers[il].attn_q_norm) {
-                    Qcur = llm_build_norm(ctx0, Qcur, hparams,
-                            model.layers[il].attn_q_norm,
-                            model.layers[il].attn_q_norm_b,
-                            LLM_NORM, cb, il);
-                }
-
-                Kcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur), model.layers[il].bk);
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
                 cb(Kcur, "Kcur", il);
 
-                if (model.layers[il].attn_k_norm) {
-                    Kcur = llm_build_norm(ctx0, Kcur, hparams,
-                            model.layers[il].attn_k_norm,
-                            model.layers[il].attn_k_norm_b,
-                            LLM_NORM, cb, il);
-                }
-                Vcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur), model.layers[il].bv);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-            } else {
-                // compute Q and K and RoPE them
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
-                Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
-                Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
                 cb(Vcur, "Vcur", il);
 
                 Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
+                        ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head_k, n_head,    n_tokens), inp_pos, nullptr,
+                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow);
                 cb(Qcur, "Qcur", il);
 
+                // ref: https://github.com/google/gemma_pytorch/commit/03e657582d17cb5a8617ebf333c1c16f3694670e
+                switch (model.type) {
+                    case llm_type::MODEL_2B:
+                    case llm_type::MODEL_9B:  Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd_head_k)));   break;
+                    case llm_type::MODEL_27B: Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd / n_head))); break;
+                    default: GGML_ABORT("fatal error");
+                };
+                cb(Qcur, "Qcur_scaled", il);
+
                 Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
+                        ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head_k, n_head_kv, n_tokens), inp_pos, nullptr,
+                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                        ext_factor, attn_factor, beta_fast, beta_slow);
                 cb(Kcur, "Kcur", il);
-            }
-
-            struct ggml_tensor * q =                 ggml_permute(ctx0, Qcur, 0, 2, 1, 3);
-            struct ggml_tensor * k = ggml_cont(ctx0, ggml_permute(ctx0, Kcur, 0, 2, 1, 3));
-
-            struct ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
-            cb(kq, "kq", il);
-
-            kq = ggml_soft_max_ext(ctx0, kq, KQ_mask, 1.0f/sqrtf(float(n_embd_head)), hparams.f_max_alibi_bias);
-            cb(kq, "kq_soft_max_ext", il);
-
-            struct ggml_tensor * v = ggml_cont(ctx0, ggml_transpose(ctx0, ggml_reshape_2d(ctx0, Vcur, n_embd_gqa, n_tokens)));
-            cb(v, "v", il);
-
-            struct ggml_tensor * kqv = ggml_mul_mat(ctx0, ggml_reshape_3d(ctx0, v, n_tokens, n_embd_head, n_head_kv), kq);
-            cb(kqv, "kqv", il);
-
-            struct ggml_tensor * kqv_merged = ggml_permute(ctx0, kqv, 0, 2, 1, 3);
-            cb(kqv_merged, "kqv_merged", il);
-
-            cur = ggml_cont_2d(ctx0, kqv_merged, n_embd_gqa, n_tokens);
-            cb(cur, "kqv_merged_cont", il);
-
-            ggml_build_forward_expand(gf, cur);
 
-            cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo, cur);
-            if (model.layers[il].bo) {
-                cb(cur, "kqv_wo", il);
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
+                        model.layers[il].wo, NULL,
+                        Kcur, Vcur, Qcur, KQ_mask_l, n_tokens, kv_head, n_kv, 1.0f, cb, il);
             }
 
-            if (model.layers[il].bo) {
-                cur = ggml_add(ctx0, cur, model.layers[il].bo);
-            }
-            cb(cur, "kqv_out", il);
+            cur = llm_build_norm(ctx0, cur, hparams,
+                    model.layers[il].attn_post_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_post_norm", il);
 
-            if (il == n_layer - 1 && pooling_type == LLAMA_POOLING_TYPE_NONE) {
+            if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
                 cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
                 inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
             }
 
-            // re-add the layer input
-            cur = ggml_add(ctx0, cur, inpL);
-
-            // attention layer norm
-            cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].attn_out_norm, model.layers[il].attn_out_norm_b, LLM_NORM, cb, il);
-
-            if (model.layers[il].attn_norm_2 != nullptr) {
-                cur = ggml_add(ctx0, cur, inpL); // re-add the layer input
-                cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].attn_norm_2, model.layers[il].attn_norm_2_b, LLM_NORM, cb, il);
-            }
+            struct ggml_tensor * sa_out = ggml_add(ctx0, cur, inpL);
+            cb(sa_out, "sa_out", il);
 
-            struct ggml_tensor * ffn_inp = cur;
-            cb(ffn_inp, "ffn_inp", il);
+            cur = llm_build_norm(ctx0, sa_out, hparams,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_norm", il);
 
             // feed-forward network
-            if (model.arch == LLM_ARCH_BERT) {
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        NULL,                      NULL,                        NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                        NULL,
-                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
-            } else if (model.arch == LLM_ARCH_JINA_BERT_V2) {
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   NULL,                        NULL,
-                        model.layers[il].ffn_gate, NULL,                        NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                        NULL,
-                        LLM_FFN_GELU, LLM_FFN_PAR, cb, il);
-            } else {
+            {
                 cur = llm_build_ffn(ctx0, lctx, cur,
                         model.layers[il].ffn_up,   NULL, NULL,
                         model.layers[il].ffn_gate, NULL, NULL,
                         model.layers[il].ffn_down, NULL, NULL,
                         NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                        LLM_FFN_GELU, LLM_FFN_PAR, cb, il);
+                cb(cur, "ffn_out", il);
             }
-            cb(cur, "ffn_out", il);
 
-            // attentions bypass the intermediate layer
-            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = llm_build_norm(ctx0, cur, hparams,
+                model.layers[il].ffn_post_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
+            cb(cur, "ffn_post_norm", -1);
 
-            // output layer norm
-            cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].layer_out_norm, model.layers[il].layer_out_norm_b, LLM_NORM, cb, il);
+            cur = ggml_add(ctx0, cur, sa_out);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
             // input for next layer
             inpL = cur;
@@ -12080,58 +7347,91 @@ struct llm_build_context {
 
         cur = inpL;
 
-        cb(cur, "result_embd", -1);
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
+        cb(cur, "result_norm", -1);
+
+        // lm_head
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+
+        // final logit soft-capping
+        cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_final_logit_softcapping);
+        cur = ggml_tanh(ctx0, cur);
+        cur = ggml_scale(ctx0, cur, hparams.f_final_logit_softcapping);
+
+        cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
 
         return gf;
     }
 
-    struct ggml_cgraph * build_bloom() {
+
+    struct ggml_cgraph * build_starcoder2() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
 
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
+        // inp_pos - contains the positions
+        struct ggml_tensor * inp_pos = build_inp_pos();
+
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
-        inpL = llm_build_norm(ctx0, inpL, hparams,
-                model.tok_norm,
-                model.tok_norm_b,
-                LLM_NORM, cb, -1);
-        cb(inpL, "inp_norm", -1);
-
         for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
+
+            // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
+                    model.layers[il].attn_norm, model.layers[il].attn_norm_b,
                     LLM_NORM, cb, il);
             cb(cur, "attn_norm", il);
 
             // self-attention
             {
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
+                // compute Q and K and RoPE them
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                cb(Qcur, "Qcur", il);
+                if (model.layers[il].bq) {
+                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                    cb(Qcur, "Qcur", il);
+                }
 
-                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                cb(cur, "bqkv", il);
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
+                if (model.layers[il].bk) {
+                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                    cb(Kcur, "Kcur", il);
+                }
 
-                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
+                if (model.layers[il].bv) {
+                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                    cb(Vcur, "Vcur", il);
+                }
 
+                Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
                 cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
+                Kcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
                         model.layers[il].wo, model.layers[il].bo,
@@ -12141,30 +7441,27 @@ struct llm_build_context {
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
             }
 
-            // Add the input
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
             cb(ffn_inp, "ffn_inp", il);
 
-            // FF
-            {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm,
-                        model.layers[il].ffn_norm_b,
-                        LLM_NORM, cb, il);
-                cb(cur, "ffn_norm", il);
+            // feed-forward network
 
-                cur = llm_build_ffn(ctx0, lctx, cur,
+            cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                    model.layers[il].ffn_norm, model.layers[il].ffn_norm_b,
+                    LLM_NORM, cb, il);
+            cb(cur, "ffn_norm", il);
+
+            cur = llm_build_ffn(ctx0, lctx, cur,
                         model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
                         NULL,                      NULL,                        NULL,
                         model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
                         NULL,
                         LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
-                cb(cur, "ffn_out", il);
-            }
+            cb(cur, "ffn_out", il);
 
             cur = ggml_add(ctx0, cur, ffn_inp);
             cur = lctx.cvec.apply_to(ctx0, cur, il);
@@ -12174,12 +7471,14 @@ struct llm_build_context {
             inpL = cur;
         }
 
-        cur = llm_build_norm(ctx0, inpL, hparams,
-                model.output_norm,
-                model.output_norm_b,
+        cur = inpL;
+
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, model.output_norm_b,
                 LLM_NORM, cb, -1);
         cb(cur, "result_norm", -1);
 
+        // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
         cb(cur, "result_output", -1);
 
@@ -12188,94 +7487,28 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_mpt() {
+    struct ggml_cgraph * build_mamba() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-
         struct ggml_tensor * cur;
-        struct ggml_tensor * pos;
         struct ggml_tensor * inpL;
 
+        // {n_embd, n_tokens}
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        if (model.pos_embd) {
-            // inp_pos - contains the positions
-            struct ggml_tensor * inp_pos = build_inp_pos();
-            pos = ggml_get_rows(ctx0, model.pos_embd, inp_pos);
-            cb(pos, "pos_embd", -1);
-
-            inpL = ggml_add(ctx0, inpL, pos);
-            cb(inpL, "inpL", -1);
-        }
+        struct ggml_tensor * state_copy = build_inp_s_copy();
+        struct ggml_tensor * state_mask = build_inp_s_mask();
 
         for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * attn_norm;
-
-            attn_norm = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, cb, il);
-            cb(attn_norm, "attn_norm", il);
-
-            // self-attention
-            {
-                cur = attn_norm;
-
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                if (model.layers[il].bqkv){
-                    cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                    cb(cur, "bqkv", il);
-                }
-
-                if (hparams.f_clamp_kqv > 0.0f) {
-                    cur = ggml_clamp(ctx0, cur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
-                    cb(cur, "wqkv_clamped", il);
-                }
-
-                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                // Q/K Layernorm
-                if (model.layers[il].attn_q_norm) {
-                    Qcur = llm_build_norm(ctx0, Qcur, hparams,
-                            model.layers[il].attn_q_norm,
-                            model.layers[il].attn_q_norm_b,
-                            LLM_NORM, cb, il);
-                    cb(Qcur, "Qcur", il);
-
-                    Kcur = llm_build_norm(ctx0, Kcur, hparams,
-                            model.layers[il].attn_k_norm,
-                            model.layers[il].attn_k_norm_b,
-                            LLM_NORM, cb, il);
-                    cb(Kcur, "Kcur", il);
-
-                    Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                    Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-
-                    cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                            model.layers[il].wo, model.layers[il].bo,
-                            Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-                } else {
-                    Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
+            // norm
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm", il);
 
-                    cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                            model.layers[il].wo, model.layers[il].bo,
-                            Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-                }
-            }
+            cur = llm_build_mamba(ctx0, lctx, ubatch, gf, cur,
+                    state_copy, state_mask,
+                    kv_head, n_kv, cb, il);
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
@@ -12284,27 +7517,8 @@ struct llm_build_context {
                 inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
             }
 
-            // Add the input
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed forward
-            {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm,
-                        model.layers[il].ffn_norm_b,
-                        LLM_NORM, cb, il);
-                cb(cur, "ffn_norm", il);
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        NULL,                      NULL,                        NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                        model.layers[il].ffn_act,
-                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
-                cb(cur, "ffn_out", il);
-            }
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
+            // residual
+            cur = ggml_add(ctx0, cur, inpL);
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
@@ -12312,14 +7526,13 @@ struct llm_build_context {
             inpL = cur;
         }
 
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm,
-                model.output_norm_b,
-                LLM_NORM, cb, -1);
+        // final rmsnorm
+        cur = llm_build_norm(ctx0, inpL, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
         cb(cur, "result_norm", -1);
 
+        // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
         cb(cur, "result_output", -1);
 
@@ -12328,11 +7541,13 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_stablelm() {
-        struct ggml_cgraph * gf = ggml_new_graph(ctx0);
+    struct ggml_cgraph * build_command_r() {
+
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        const float f_logit_scale = hparams.f_logit_scale;
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
@@ -12347,15 +7562,12 @@ struct llm_build_context {
 
         for (int il = 0; il < n_layer; ++il) {
 
-
             // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
+                    model.layers[il].attn_norm, NULL,
                     LLM_NORM, cb, il);
             cb(cur, "attn_norm", il);
-
-            struct ggml_tensor * inpSA = cur;
+            struct ggml_tensor * ffn_inp = cur;
 
             // self-attention
             {
@@ -12381,19 +7593,24 @@ struct llm_build_context {
                     cb(Vcur, "Vcur", il);
                 }
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                cb(Qcur, "Qcur", il);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-                cb(Kcur, "Kcur", il);
-
                 if (model.layers[il].attn_q_norm) {
+                    Qcur = ggml_view_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens,
+                                ggml_element_size(Qcur) * n_embd_head,
+                                ggml_element_size(Qcur) * n_embd_head * n_head,
+                                0);
+                    cb(Qcur, "Qcur", il);
+                    Kcur = ggml_view_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens,
+                                ggml_element_size(Kcur) * n_embd_head,
+                                ggml_element_size(Kcur) * n_embd_head * n_head_kv,
+                                0);
+                    cb(Kcur, "Kcur", il);
+
                     Qcur = llm_build_norm(ctx0, Qcur, hparams,
-                            model.layers[il].attn_q_norm,
-                            NULL,
-                            LLM_NORM, cb, il);
+                                model.layers[il].attn_q_norm,
+                                NULL,
+                                LLM_NORM, cb, il);
                     cb(Qcur, "Qcur", il);
-                }
-                if (model.layers[il].attn_k_norm) {
+
                     Kcur = llm_build_norm(ctx0, Kcur, hparams,
                             model.layers[il].attn_k_norm,
                             NULL,
@@ -12401,50 +7618,38 @@ struct llm_build_context {
                     cb(Kcur, "Kcur", il);
                 }
 
-
                 Qcur = ggml_rope_ext(
-                    ctx0, Qcur, inp_pos, nullptr,
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
                 cb(Qcur, "Qcur", il);
 
                 Kcur = ggml_rope_ext(
-                    ctx0, Kcur, inp_pos, nullptr,
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
                 cb(Kcur, "Kcur", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
+                        model.layers[il].wo, model.layers[il].bo,
                         Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpL  = ggml_get_rows(ctx0,  inpL, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+                cur     = ggml_get_rows(ctx0,     cur, inp_out_ids);
+                inpL    = ggml_get_rows(ctx0,    inpL, inp_out_ids);
+                ffn_inp = ggml_get_rows(ctx0, ffn_inp, inp_out_ids);
             }
 
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
-            cb(ffn_inp, "ffn_inp", il);
+            struct ggml_tensor * attn_out = cur;
 
             // feed-forward network
             {
-                if (model.layers[il].ffn_norm) {
-                    cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                            model.layers[il].ffn_norm,
-                            model.layers[il].ffn_norm_b,
-                            LLM_NORM, cb, il);
-                    cb(cur, "ffn_norm", il);
-                } else {
-                    // parallel residual
-                    cur = inpSA;
-                }
-                cur = llm_build_ffn(ctx0, lctx, cur,
+                cur = llm_build_ffn(ctx0, lctx, ffn_inp,
                         model.layers[il].ffn_up,   NULL, NULL,
                         model.layers[il].ffn_gate, NULL, NULL,
                         model.layers[il].ffn_down, NULL, NULL,
@@ -12453,7 +7658,9 @@ struct llm_build_context {
                 cb(cur, "ffn_out", il);
             }
 
-            cur = ggml_add(ctx0, cur, ffn_inp);
+            // add together residual + FFN + self-attention
+            cur = ggml_add(ctx0, cur, inpL);
+            cur = ggml_add(ctx0, cur, attn_out);
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
@@ -12464,25 +7671,31 @@ struct llm_build_context {
         cur = inpL;
 
         cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm,
-                model.output_norm_b,
+                model.output_norm, NULL,
                 LLM_NORM, cb, -1);
         cb(cur, "result_norm", -1);
 
         // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+
+        if (f_logit_scale) {
+            cur = ggml_scale(ctx0, cur, f_logit_scale);
+        }
+
         cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
 
         return gf;
+
     }
 
-    struct ggml_cgraph * build_qwen() {
+    struct ggml_cgraph * build_cohere2() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        const float f_logit_scale = hparams.f_logit_scale;
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
@@ -12493,80 +7706,95 @@ struct llm_build_context {
         struct ggml_tensor * inp_pos = build_inp_pos();
 
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
+        // cohere2 requires different mask for layers using sliding window (SWA)
+        struct ggml_tensor * KQ_mask     = build_inp_KQ_mask();
+        struct ggml_tensor * KQ_mask_swa = build_inp_KQ_mask_swa();
+
+        // sliding window switch pattern
+        const int32_t sliding_window_pattern = 4;
 
         for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
+            // three layers sliding window attention (window size 4096) and ROPE
+            // fourth layer uses global attention without positional embeddings
+            const bool           is_sliding = il % sliding_window_pattern < (sliding_window_pattern - 1);
+            struct ggml_tensor * KQ_mask_l = is_sliding ? KQ_mask_swa : KQ_mask;
 
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
+            // norm
+            cur = llm_build_norm(ctx0, inpL, hparams, model.layers[il].attn_norm, NULL, LLM_NORM, cb, il);
             cb(cur, "attn_norm", il);
+            struct ggml_tensor * ffn_inp = cur;
 
             // self-attention
             {
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                cb(cur, "bqkv", il);
-
-                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd, n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd, n_tokens, cur->nb[1], 2*sizeof(float)*(n_embd)));
+                // rope freq factors for 128k context
+                struct ggml_tensor * rope_factors = build_rope_factors(il);
 
+                // compute Q and K and RoPE them
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
                 cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
+                if (model.layers[il].bq) {
+                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                    cb(Qcur, "Qcur", il);
+                }
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
+                if (model.layers[il].bk) {
+                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                    cb(Kcur, "Kcur", il);
+                }
 
-                // using mode = 2 for neox mode
-                Qcur = ggml_rope_ext(
-                    ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig,
-                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur", il);
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
+                if (model.layers[il].bv) {
+                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                    cb(Vcur, "Vcur", il);
+                }
 
-                Kcur = ggml_rope_ext(
-                    ctx0, Kcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig,
-                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur", il);
+                if (is_sliding) {
+                    Qcur = ggml_rope_ext(ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, rope_factors,
+                                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale, ext_factor, attn_factor,
+                                        beta_fast, beta_slow);
+                    cb(Qcur, "Qcur", il);
 
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+                    Kcur = ggml_rope_ext(ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos,
+                                        rope_factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale, ext_factor,
+                                        attn_factor, beta_fast, beta_slow);
+                    cb(Kcur, "Kcur", il);
+                } else {
+                    // For non-sliding layers, just reshape without applying RoPE
+                    Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
+                    cb(Qcur, "Qcur", il);
+
+                    Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+                    cb(Kcur, "Kcur", il);
+                }
+
+                cur = llm_build_kv(ctx0, lctx, kv_self, gf, model.layers[il].wo, model.layers[il].bo, Kcur, Vcur, Qcur,
+                                   KQ_mask_l, n_tokens, kv_head, n_kv, 1.0f / sqrtf(float(n_embd_head)), cb, il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+                cur                              = ggml_get_rows(ctx0, cur, inp_out_ids);
+                inpL                             = ggml_get_rows(ctx0, inpL, inp_out_ids);
+                ffn_inp                          = ggml_get_rows(ctx0, ffn_inp, inp_out_ids);
             }
 
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            struct ggml_tensor * attn_out = cur;
 
-            // feed-forward forward
+            // feed-forward network
             {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                cur = llm_build_ffn(ctx0, lctx, ffn_inp, model.layers[il].ffn_up, NULL, NULL, model.layers[il].ffn_gate,
+                                    NULL, NULL, model.layers[il].ffn_down, NULL, NULL, NULL, LLM_FFN_SILU, LLM_FFN_PAR,
+                                    cb, il);
                 cb(cur, "ffn_out", il);
             }
 
-            cur = ggml_add(ctx0, cur, ffn_inp);
+            // add together residual + FFN + self-attention
+            cur = ggml_add(ctx0, cur, inpL);
+            cur = ggml_add(ctx0, cur, attn_out);
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
@@ -12576,13 +7804,16 @@ struct llm_build_context {
 
         cur = inpL;
 
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
+        cur = llm_build_norm(ctx0, cur, hparams, model.output_norm, NULL, LLM_NORM, cb, -1);
         cb(cur, "result_norm", -1);
 
         // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+
+        if (f_logit_scale) {
+            cur = ggml_scale(ctx0, cur, f_logit_scale);
+        }
+
         cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
@@ -12590,9 +7821,18 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_qwen2() {
+    // ref: https://allenai.org/olmo
+    // based on the original build_llama() function, changes:
+    //   * non-parametric layer norm
+    //   * clamp qkv
+    //   * removed bias
+    //   * removed MoE
+    struct ggml_cgraph * build_olmo() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
+
         const int64_t n_embd_head = hparams.n_embd_head_v;
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
         GGML_ASSERT(n_embd_head == hparams.n_rot);
@@ -12613,8 +7853,8 @@ struct llm_build_context {
 
             // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
+                    NULL, NULL,
+                    LLM_NORM, cb, il);
             cb(cur, "attn_norm", il);
 
             // self-attention
@@ -12622,21 +7862,27 @@ struct llm_build_context {
                 // compute Q and K and RoPE them
                 struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
                 cb(Qcur, "Qcur", il);
-                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                cb(Qcur, "Qcur", il);
+                if (hparams.f_clamp_kqv > 0.0f) {
+                    Qcur = ggml_clamp(ctx0, Qcur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
+                    cb(Qcur, "Qcur", il);
+                }
 
                 struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
                 cb(Kcur, "Kcur", il);
-                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                cb(Kcur, "Kcur", il);
+                if (hparams.f_clamp_kqv > 0.0f) {
+                    Kcur = ggml_clamp(ctx0, Kcur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
+                    cb(Kcur, "Kcur", il);
+                }
 
                 struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
                 cb(Vcur, "Vcur", il);
-                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                cb(Vcur, "Vcur", il);
+                if (hparams.f_clamp_kqv > 0.0f) {
+                    Vcur = ggml_clamp(ctx0, Vcur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
+                    cb(Vcur, "Vcur", il);
+                }
 
                 Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens), inp_pos, nullptr,
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
@@ -12650,13 +7896,14 @@ struct llm_build_context {
                 cb(Kcur, "Kcur", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
+                        model.layers[il].wo, nullptr,
                         Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                n_tokens = n_outputs;
                 cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
                 inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
             }
@@ -12666,8 +7913,8 @@ struct llm_build_context {
 
             // feed-forward network
             cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
+                    NULL, NULL,
+                    LLM_NORM, cb, il);
             cb(cur, "ffn_norm", il);
 
             cur = llm_build_ffn(ctx0, lctx, cur,
@@ -12679,6 +7926,8 @@ struct llm_build_context {
             cb(cur, "ffn_out", il);
 
             cur = ggml_add(ctx0, cur, ffn_inp);
+            cb(cur, "ffn_out", il);
+
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
@@ -12689,8 +7938,8 @@ struct llm_build_context {
         cur = inpL;
 
         cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
+                NULL, NULL,
+                LLM_NORM, cb, -1);
         cb(cur, "result_norm", -1);
 
         // lm_head
@@ -12702,8 +7951,12 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_qwen2vl() {
+    struct ggml_cgraph * build_olmo2() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
+
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
+
         const int64_t n_embd_head = hparams.n_embd_head_v;
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
         GGML_ASSERT(n_embd_head == hparams.n_rot);
@@ -12714,67 +7967,67 @@ struct llm_build_context {
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
         // inp_pos - contains the positions
-        lctx.inp_pos = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_tokens * 4);
-        cb(lctx.inp_pos, "inp_pos", -1);
-        ggml_set_input(lctx.inp_pos);
-        struct ggml_tensor * inp_pos = lctx.inp_pos;
+        struct ggml_tensor * inp_pos = build_inp_pos();
 
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-        int sections[4];
-        std::copy(std::begin(hparams.rope_sections), std::begin(hparams.rope_sections) + 4, sections);
 
         for (int il = 0; il < n_layer; ++il) {
             struct ggml_tensor * inpSA = inpL;
 
-            // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
+            cur = inpL;
 
-            // self-attention
+            // self_attention
             {
                 // compute Q and K and RoPE them
                 struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
                 cb(Qcur, "Qcur", il);
-                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                cb(Qcur, "Qcur", il);
 
                 struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
                 cb(Kcur, "Kcur", il);
-                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                cb(Kcur, "Kcur", il);
 
                 struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
                 cb(Vcur, "Vcur", il);
-                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_rope_multi(
-                    ctx0,
-                    ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
-                    n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale,
+                Qcur = llm_build_norm(ctx0, Qcur, hparams, model.layers[il].attn_q_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(Qcur, "Qcur_normed", il);
+
+                Kcur = llm_build_norm(ctx0, Kcur, hparams, model.layers[il].attn_k_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(Kcur, "Kcur_normed", il);
+
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
+                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+
+                Qcur = ggml_rope_ext(
+                    ctx0, Qcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
-                cb(Qcur, "Qcur", il);
+                cb(Qcur, "Qcur_rope", il);
 
-                Kcur = ggml_rope_multi(
-                    ctx0,
-                    ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
-                    n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale,
+                Kcur = ggml_rope_ext(
+                    ctx0, Kcur, inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
-                cb(Kcur, "Kcur", il);
+                cb(Kcur, "Kcur_rope", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
+                        model.layers[il].wo, NULL,
                         Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
             }
 
+            cur = llm_build_norm(ctx0, cur, hparams,
+                    model.layers[il].attn_post_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_post_norm", il);
+
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                n_tokens = n_outputs;
                 cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
                 inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
             }
@@ -12783,12 +8036,7 @@ struct llm_build_context {
             cb(ffn_inp, "ffn_inp", il);
 
             // feed-forward network
-            cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = llm_build_ffn(ctx0, lctx, cur,
+            cur = llm_build_ffn(ctx0, lctx, ffn_inp,
                     model.layers[il].ffn_up,   NULL, NULL,
                     model.layers[il].ffn_gate, NULL, NULL,
                     model.layers[il].ffn_down, NULL, NULL,
@@ -12796,7 +8044,14 @@ struct llm_build_context {
                     LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
             cb(cur, "ffn_out", il);
 
+            cur = llm_build_norm(ctx0, cur, hparams,
+                model.layers[il].ffn_post_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
+            cb(cur, "ffn_post_norm", -1);
+
             cur = ggml_add(ctx0, cur, ffn_inp);
+            cb(cur, "ffn_out", il);
+
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
@@ -12820,7 +8075,11 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_qwen2moe() {
+    // based on the build_qwen2moe() function, changes:
+    //   * removed shared experts
+    //   * removed bias
+    //   * added q, k norm
+    struct ggml_cgraph * build_olmoe() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
         // mutable variable, needed during the last layer of the computation to skip unused tokens
@@ -12855,35 +8114,40 @@ struct llm_build_context {
                 // compute Q and K and RoPE them
                 struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
                 cb(Qcur, "Qcur", il);
-                Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                cb(Qcur, "Qcur", il);
 
                 struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
                 cb(Kcur, "Kcur", il);
-                Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                cb(Kcur, "Kcur", il);
 
                 struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
                 cb(Vcur, "Vcur", il);
-                Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                cb(Vcur, "Vcur", il);
+
+                Qcur = llm_build_norm(ctx0, Qcur, hparams, model.layers[il].attn_q_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(Qcur, "Qcur_normed", il);
+
+                Kcur = llm_build_norm(ctx0, Kcur, hparams, model.layers[il].attn_k_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(Kcur, "Kcur_normed", il);
+
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
+                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
 
                 Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                    ctx0, Qcur, inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
-                cb(Qcur, "Qcur", il);
+                cb(Qcur, "Qcur_rope", il);
 
                 Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    ctx0, Kcur, inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
-                cb(Kcur, "Kcur", il);
+                cb(Kcur, "Kcur_rope", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
+                        model.layers[il].wo, NULL,
                         Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
             }
 
@@ -12904,47 +8168,22 @@ struct llm_build_context {
                     LLM_NORM_RMS, cb, il);
             cb(cur, "ffn_norm", il);
 
-            ggml_tensor * moe_out =
-                    llm_build_moe_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_gate_inp,
-                        model.layers[il].ffn_up_exps,
-                        model.layers[il].ffn_gate_exps,
-                        model.layers[il].ffn_down_exps,
-                        n_expert, n_expert_used,
-                        LLM_FFN_SILU, false,
-                        false, 0.0,
-                        cb, il);
+            cur = llm_build_moe_ffn(ctx0, lctx, cur,
+                    model.layers[il].ffn_gate_inp,
+                    model.layers[il].ffn_up_exps,
+                    model.layers[il].ffn_gate_exps,
+                    model.layers[il].ffn_down_exps,
+                    nullptr,
+                    n_expert, n_expert_used,
+                    LLM_FFN_SILU, false,
+                    false, 0.0,
+                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                    cb, il);
             cb(cur, "ffn_moe_out", il);
 
-            // FFN shared expert
-            {
-                ggml_tensor * cur_gate_inp = llm_build_lora_mm(lctx, ctx0, model.layers[il].ffn_gate_inp_shexp, cur);
-                cb(cur_gate_inp, "ffn_shexp_gate_inp", il);
-
-                // sigmoid
-                ggml_tensor * cur_gate = ggml_div(ctx0, ggml_silu(ctx0, cur_gate_inp), cur_gate_inp);
-                cb(cur_gate, "ffn_shexp_gate", il);
-
-                ggml_tensor * cur_ffn = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up_shexp,   NULL, NULL,
-                        model.layers[il].ffn_gate_shexp, NULL, NULL,
-                        model.layers[il].ffn_down_shexp, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-                cb(cur_ffn, "ffn_shexp", il);
-
-                ggml_tensor * ffn_shexp_out = ggml_mul(ctx0, cur_ffn, cur_gate);
-                cb(ffn_shexp_out, "ffn_shexp_out", il);
-
-                moe_out = ggml_add(ctx0, moe_out, ffn_shexp_out);
-                cb(moe_out, "ffn_out", il);
-
-                cur = moe_out;
-            }
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
 
             // input for next layer
             inpL = cur;
@@ -12966,18 +8205,14 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_phi2() {
+    struct ggml_cgraph * build_openelm() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
         struct ggml_tensor * cur;
-        struct ggml_tensor * attn_norm_output;
-        struct ggml_tensor * ffn_output;
         struct ggml_tensor * inpL;
-
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
         // inp_pos - contains the positions
@@ -12987,111 +8222,119 @@ struct llm_build_context {
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
         for (int il = 0; il < n_layer; ++il) {
-            attn_norm_output = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, cb, il);
-            cb(attn_norm_output, "attn_norm", il);
+            const int64_t n_head    = hparams.n_head(il);
+            const int64_t n_head_kv = hparams.n_head_kv(il);
+            const int64_t n_head_qkv = 2*n_head_kv + n_head;
 
-            // self-attention
-            {
-                struct ggml_tensor * Qcur = nullptr;
-                struct ggml_tensor * Kcur = nullptr;
-                struct ggml_tensor * Vcur = nullptr;
+            cur = inpL;
+            struct ggml_tensor * residual = cur;
 
-                if (model.layers[il].wqkv) {
-                    cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, attn_norm_output);
-                    cb(cur, "wqkv", il);
+            // norm
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm", il);
 
-                    cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                    cb(cur, "bqkv", il);
+            // self-attention
+            {
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-                    Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
-                    Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
-                    Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
-                } else {
-                    Qcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, attn_norm_output), model.layers[il].bq);
-                    Kcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, attn_norm_output), model.layers[il].bk);
-                    Vcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, attn_norm_output), model.layers[il].bv);
-                }
+                cur = ggml_reshape_3d(ctx0, cur, n_embd_head_k, n_head_qkv, n_tokens);
 
+                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd_head, n_head, n_tokens, cur->nb[1], cur->nb[2], 0));
                 cb(Qcur, "Qcur", il);
+
+                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, cur->nb[1], cur->nb[2], cur->nb[1]*n_head));
                 cb(Kcur, "Kcur", il);
+
+                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, cur->nb[1], cur->nb[2], cur->nb[1]*(n_head+n_head_kv)));
                 cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+                Qcur = llm_build_norm(ctx0, Qcur, hparams,
+                        model.layers[il].attn_q_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(Qcur, "Qcur", il);
+
+                Kcur = llm_build_norm(ctx0, Kcur, hparams,
+                        model.layers[il].attn_k_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(Kcur, "Kcur", il);
 
                 Qcur = ggml_rope_ext(
-                    ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig,
+                    ctx0, Qcur, inp_pos, NULL, n_rot, rope_type, n_ctx_orig,
                     freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
                 );
                 cb(Qcur, "Qcur", il);
 
-                // with phi2, we scale the Q to avoid precision issues
-                // ref: https://github.com/ml-explore/mlx-examples/blob/08e862336ade809bc37d1035f94b359e7d1a5152/phi2/phi2.py#L64-L66
-                Qcur = ggml_scale(ctx0, Qcur, 1.0f/sqrtf(float(n_embd_head)));
-                cb(Qcur, "Qcur", il);
-
                 Kcur = ggml_rope_ext(
-                    ctx0, Kcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig,
+                    ctx0, Kcur, inp_pos, NULL, n_rot, rope_type, n_ctx_orig,
                     freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
                 );
                 cb(Kcur, "Kcur", il);
 
+                Vcur = ggml_reshape_2d(ctx0, Vcur, n_embd_head * n_head_kv, n_tokens);
+                cb(Qcur, "Vcur", il);
+
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f, cb, il);
+                        model.layers[il].wo, NULL,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur              = ggml_get_rows(ctx0,              cur, inp_out_ids);
-                inpL             = ggml_get_rows(ctx0,             inpL, inp_out_ids);
-                attn_norm_output = ggml_get_rows(ctx0, attn_norm_output, inp_out_ids);
+                residual = ggml_get_rows(ctx0, residual, inp_out_ids);
+                cur = ggml_get_rows(ctx0, cur, inp_out_ids);
             }
 
-            // FF
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, residual, cur);
+            cb(ffn_inp, "ffn_inp", il);
+
+            // feed-forward network
             {
-                ffn_output = llm_build_ffn(ctx0, lctx, attn_norm_output,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        NULL,                      NULL,                        NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "ffn_norm", il);
+
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        model.layers[il].ffn_gate, NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
                         NULL,
-                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
-                cb(ffn_output, "ffn_out", il);
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                cb(cur, "ffn_out", il);
             }
 
-            cur = ggml_add(ctx0, cur, ffn_output);
-            cur = ggml_add(ctx0, cur, inpL);
+            cur = ggml_add(ctx0, cur, ffn_inp);
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
-            // input for next layer
             inpL = cur;
         }
 
-        cur = llm_build_norm(ctx0, inpL, hparams,
-                model.output_norm,
-                model.output_norm_b,
-                LLM_NORM, cb, -1);
+        cur = inpL;
+
+        // norm
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
         cb(cur, "result_norm", -1);
 
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output_no_bias", -1);
-
-        cur = ggml_add(ctx0, cur, model.output_b);
         cb(cur, "result_output", -1);
+
         ggml_build_forward_expand(gf, cur);
+
         return gf;
     }
 
-    struct ggml_cgraph * build_phi3() {
+    struct ggml_cgraph * build_gptneox() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
         struct ggml_tensor * cur;
@@ -13103,108 +8346,123 @@ struct llm_build_context {
         struct ggml_tensor * inp_pos = build_inp_pos();
 
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask_swa = build_inp_KQ_mask_swa();
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
         for (int il = 0; il < n_layer; ++il) {
-            auto residual = inpL;
+            cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm,
+                    model.layers[il].attn_norm_b,
+                    LLM_NORM, cb, il);
+            cb(cur, "attn_norm", il);
 
             // self-attention
             {
-                // rope freq factors for 128k context
-                struct ggml_tensor * rope_factors = build_rope_factors(il);
-
-                struct ggml_tensor* attn_norm_output = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    NULL,
-                    LLM_NORM_RMS, cb, il);
-                cb(attn_norm_output, "attn_norm", il);
-
-                struct ggml_tensor * Qcur = nullptr;
-                struct ggml_tensor * Kcur = nullptr;
-                struct ggml_tensor * Vcur = nullptr;
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-                if (model.layers[il].wqkv) {
-                    cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, attn_norm_output);
-                    cb(cur, "wqkv", il);
+                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
+                cb(cur, "bqkv", il);
 
-                    Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0 * sizeof(float) * (n_embd)));
-                    Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1 * sizeof(float) * (n_embd)));
-                    Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1 * sizeof(float) * (n_embd + n_embd_gqa)));
-                }
-                else {
-                    Qcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, attn_norm_output), model.layers[il].bq);
-                    Kcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, attn_norm_output), model.layers[il].bk);
-                    Vcur = ggml_add(ctx0, llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, attn_norm_output), model.layers[il].bv);
-                }
+                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
+                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
 
                 cb(Qcur, "Qcur", il);
                 cb(Kcur, "Kcur", il);
                 cb(Vcur, "Vcur", il);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-
                 Qcur = ggml_rope_ext(
-                    ctx0, Qcur, inp_pos, rope_factors, n_rot, rope_type, n_ctx_orig,
-                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
                 );
                 cb(Qcur, "Qcur", il);
 
-                Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd_head)));
-                cb(Qcur, "Qcur", il);
-
                 Kcur = ggml_rope_ext(
-                    ctx0, Kcur, inp_pos, rope_factors, n_rot, rope_type, n_ctx_orig,
-                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
                 );
                 cb(Kcur, "Kcur", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
                         model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask_swa, n_tokens, kv_head, n_kv, 1.0f, cb, il);
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
-                struct ggml_tensor* inp_out_ids = build_inp_out_ids();
-                cur = ggml_get_rows(ctx0, cur, inp_out_ids);
-                residual = ggml_get_rows(ctx0, residual, inp_out_ids);
+                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
+                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
             }
 
-            cur = ggml_add(ctx0, cur, residual);
-            residual = cur;
+            // ffn
+            if (hparams.use_par_res) {
+                // attention and ffn are computed in parallel
+                // x = x + attn(ln1(x)) + ffn(ln2(x))
 
-            cur = llm_build_norm(ctx0, cur, hparams,
-                model.layers[il].ffn_norm, NULL,
-                LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm", il);
+                struct ggml_tensor * attn_out = cur;
+
+                cur = llm_build_norm(ctx0, inpL, hparams,
+                        model.layers[il].ffn_norm,
+                        model.layers[il].ffn_norm_b,
+                        LLM_NORM, cb, il);
+                cb(cur, "ffn_norm", il);
 
-            // FF
-            // special-case: the up and gate tensors are merged into a single tensor
-            // TOOD: support into llm_build_ffn
-            {
                 cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        NULL,                      NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
+                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                        NULL,                      NULL,                        NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
                         NULL,
-                        LLM_FFN_SWIGLU, LLM_FFN_SEQ, cb, il);
+                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
                 cb(cur, "ffn_out", il);
-            }
-
-            cur = ggml_add(ctx0, residual, cur);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
 
-            // input for next layer
-            inpL = cur;
-        }
+                cur = ggml_add(ctx0, cur, inpL);
+                cb(cur, "ffn_out", il);
 
-        cur = llm_build_norm(ctx0, inpL, hparams,
-            model.output_norm,
-            NULL,
-            LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
+                cur = ggml_add(ctx0, cur, attn_out);
+                cur = lctx.cvec.apply_to(ctx0, cur, il);
+                cb(cur, "l_out", il);
+
+                // input for next layer
+                inpL = cur;
+            } else {
+                // attention and ffn are computed sequentially
+                // x = x + attn(ln1(x))
+                // x = x + ffn(ln2(x))
+
+                struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
+                cb(ffn_inp, "ffn_inp", il);
+
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm,
+                        model.layers[il].ffn_norm_b,
+                        LLM_NORM, cb, il);
+                cb(cur, "ffn_norm", il);
+
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                        NULL,                      NULL,                        NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        NULL,
+                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
+                cb(cur, "ffn_out", il);
+
+                cur = ggml_add(ctx0, cur, ffn_inp);
+                cur = lctx.cvec.apply_to(ctx0, cur, il);
+                cb(cur, "l_out", il);
+
+                // input for next layer
+                inpL = cur;
+            }
+        }
+
+        cur = llm_build_norm(ctx0, inpL, hparams,
+                model.output_norm,
+                model.output_norm_b,
+                LLM_NORM, cb, -1);
+        cb(cur, "result_norm", -1);
 
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
         cb(cur, "result_output", -1);
@@ -13214,9 +8472,11 @@ struct llm_build_context {
         return gf;
     }
 
+    struct ggml_cgraph * build_arctic() {
+        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-    struct ggml_cgraph * build_plamo() {
-        struct ggml_cgraph * gf = ggml_new_graph(ctx0);
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@@ -13234,6 +8494,7 @@ struct llm_build_context {
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
         for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
 
             // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
@@ -13241,8 +8502,6 @@ struct llm_build_context {
                     LLM_NORM_RMS, cb, il);
             cb(cur, "attn_norm", il);
 
-            struct ggml_tensor * attention_norm = cur;
-
             // self-attention
             {
                 // compute Q and K and RoPE them
@@ -13256,46 +8515,74 @@ struct llm_build_context {
                 cb(Vcur, "Vcur", il);
 
                 Qcur = ggml_rope_ext(
-                        ctx0, ggml_reshape_3d(ctx0, Qcur, n_rot, n_head,    n_tokens), inp_pos, nullptr,
-                        n_embd_head, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
                 cb(Qcur, "Qcur", il);
 
                 Kcur = ggml_rope_ext(
-                        ctx0, ggml_reshape_3d(ctx0, Kcur, n_rot, n_head_kv, n_tokens), inp_pos, nullptr,
-                        n_embd_head, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
                 cb(Kcur, "Kcur", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
                         model.layers[il].wo, NULL,
                         Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
             }
-            struct ggml_tensor * sa_out = cur;
-
-            cur = attention_norm;
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur    = ggml_get_rows(ctx0,    cur, inp_out_ids);
-                sa_out = ggml_get_rows(ctx0, sa_out, inp_out_ids);
-                inpL   = ggml_get_rows(ctx0,   inpL, inp_out_ids);
+                n_tokens = n_outputs;
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
             }
 
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
+
             // feed-forward network
-            {
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-                cb(cur, "ffn_out", il);
-            }
+            cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_norm", il);
+
+            cur = llm_build_ffn(ctx0, lctx, cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+            cb(cur, "ffn_out", il);
+
+            struct ggml_tensor * ffn_out = ggml_add(ctx0, cur, ffn_inp);
+            cb(ffn_out, "ffn_out", il);
+
+            // MoE
+            cur = llm_build_norm(ctx0, inpSA, hparams,
+                    model.layers[il].ffn_norm_exps, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_norm_exps", il);
+
+            cur = llm_build_moe_ffn(ctx0, lctx, cur,
+                    model.layers[il].ffn_gate_inp,
+                    model.layers[il].ffn_up_exps,
+                    model.layers[il].ffn_gate_exps,
+                    model.layers[il].ffn_down_exps,
+                    nullptr,
+                    n_expert, n_expert_used,
+                    LLM_FFN_SILU, true,
+                    false, 0.0,
+                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                    cb, il);
+            cb(cur, "ffn_moe_out", il);
+
+            cur = ggml_add(ctx0, cur, ffn_out);
+            cb(cur, "ffn_out", il);
 
-            cur = ggml_add(ctx0, cur, sa_out);
-            cur = ggml_add(ctx0, cur, inpL);
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
@@ -13319,15 +8606,17 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_gpt2() {
+    struct ggml_cgraph * build_deepseek() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
+
         const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
         struct ggml_tensor * cur;
-        struct ggml_tensor * pos;
         struct ggml_tensor * inpL;
 
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
@@ -13337,69 +8626,116 @@ struct llm_build_context {
 
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        pos = ggml_get_rows(ctx0, model.pos_embd, inp_pos);
-        cb(pos, "pos_embd", -1);
-
-        inpL = ggml_add(ctx0, inpL, pos);
-        cb(inpL, "inpL", -1);
-
+        const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
         for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
+
+            // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, cb, il);
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
             cb(cur, "attn_norm", il);
 
             // self-attention
             {
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                cb(cur, "bqkv", il);
-
-                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
+                // rope freq factors for llama3; may return nullptr for llama2 and other models
+                struct ggml_tensor * rope_factors = build_rope_factors(il);
 
+                // compute Q and K and RoPE them
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
                 cb(Qcur, "Qcur", il);
+                if (model.layers[il].bq) {
+                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                    cb(Qcur, "Qcur", il);
+                }
+
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
                 cb(Kcur, "Kcur", il);
+                if (model.layers[il].bk) {
+                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                    cb(Kcur, "Kcur", il);
+                }
+
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
                 cb(Vcur, "Vcur", il);
+                if (model.layers[il].bv) {
+                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                    cb(Vcur, "Vcur", il);
+                }
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
+                Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur", il);
+
+                Kcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, rope_factors,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
                         model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, kq_scale, cb, il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+                n_tokens = n_outputs;
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
             }
 
-            // add the input
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
+
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
             cb(ffn_inp, "ffn_inp", il);
 
-            // FF
-            {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm,
-                        model.layers[il].ffn_norm_b,
-                        LLM_NORM, cb, il);
-                cb(cur, "ffn_norm", il);
+            cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_norm", il);
 
+            if ((uint32_t) il < hparams.n_layer_dense_lead) {
                 cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        NULL,                      NULL,                        NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        model.layers[il].ffn_gate, NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
                         NULL,
-                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
                 cb(cur, "ffn_out", il);
+            } else {
+                // MoE branch
+                ggml_tensor * moe_out =
+                        llm_build_moe_ffn(ctx0, lctx, cur,
+                            model.layers[il].ffn_gate_inp,
+                            model.layers[il].ffn_up_exps,
+                            model.layers[il].ffn_gate_exps,
+                            model.layers[il].ffn_down_exps,
+                            nullptr,
+                            n_expert, n_expert_used,
+                            LLM_FFN_SILU, false,
+                            false, hparams.expert_weights_scale,
+                            LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                            cb, il);
+                cb(moe_out, "ffn_moe_out", il);
+
+                // FFN shared expert
+                {
+                    ggml_tensor * ffn_shexp = llm_build_ffn(ctx0, lctx, cur,
+                            model.layers[il].ffn_up_shexp,   NULL, NULL,
+                            model.layers[il].ffn_gate_shexp, NULL, NULL,
+                            model.layers[il].ffn_down_shexp, NULL, NULL,
+                            NULL,
+                            LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                    cb(ffn_shexp, "ffn_shexp", il);
+
+                    cur = ggml_add(ctx0, moe_out, ffn_shexp);
+                    cb(cur, "ffn_out", il);
+                }
             }
 
             cur = ggml_add(ctx0, cur, ffn_inp);
@@ -13410,13 +8746,16 @@ struct llm_build_context {
             inpL = cur;
         }
 
-        cur = llm_build_norm(ctx0, inpL, hparams,
-                model.output_norm,
-                model.output_norm_b,
-                LLM_NORM, cb, -1);
+        cur = inpL;
+
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
         cb(cur, "result_norm", -1);
 
+        // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+
         cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
@@ -13424,17 +8763,28 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_codeshell() {
+    struct ggml_cgraph * build_deepseek2() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
+
+        bool is_lite = (hparams.n_layer == 27);
+
+        // We have to pre-scale kq_scale and attn_factor to make the YaRN RoPE work correctly.
+        // See https://github.com/ggerganov/llama.cpp/discussions/7416 for detailed explanation.
+        const float mscale = attn_factor * (1.0f + hparams.rope_yarn_log_mul * logf(1.0f / freq_scale));
+        const float kq_scale = 1.0f*mscale*mscale/sqrtf(float(hparams.n_embd_head_k));
+        const float attn_factor_scaled = 1.0f / (1.0f + 0.1f * logf(1.0f / freq_scale));
+
+        const uint32_t n_embd_head_qk_rope = hparams.n_rot;
+        const uint32_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot;
+        const uint32_t kv_lora_rank = hparams.n_lora_kv;
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
 
+        // {n_embd, n_tokens}
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
         // inp_pos - contains the positions
@@ -13444,73 +8794,179 @@ struct llm_build_context {
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
         for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
+
+            // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, cb, il);
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
             cb(cur, "attn_norm", il);
 
-            // self-attention
+            // self_attention
             {
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
+                struct ggml_tensor * q = NULL;
+                if (!is_lite) {
+                    // {n_embd, q_lora_rank} * {n_embd, n_tokens} -> {q_lora_rank, n_tokens}
+                    q = ggml_mul_mat(ctx0, model.layers[il].wq_a, cur);
+                    cb(q, "q", il);
 
-                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                cb(cur, "bqkv", il);
+                    q = llm_build_norm(ctx0, q, hparams,
+                            model.layers[il].attn_q_a_norm, NULL,
+                            LLM_NORM_RMS, cb, il);
+                    cb(q, "q", il);
 
-                struct ggml_tensor * tmpq = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
-                struct ggml_tensor * tmpk = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
+                    // {q_lora_rank, n_head * hparams.n_embd_head_k} * {q_lora_rank, n_tokens} -> {n_head * hparams.n_embd_head_k, n_tokens}
+                    q = ggml_mul_mat(ctx0, model.layers[il].wq_b, q);
+                    cb(q, "q", il);
+                } else {
+                    q = ggml_mul_mat(ctx0, model.layers[il].wq, cur);
+                    cb(q, "q", il);
+                }
 
-                cb(tmpq, "tmpq", il);
-                cb(tmpk, "tmpk", il);
-                cb(Vcur, "Vcur", il);
+                // split into {n_head * n_embd_head_qk_nope, n_tokens}
+                struct ggml_tensor * q_nope = ggml_view_3d(ctx0, q, n_embd_head_qk_nope, n_head, n_tokens,
+                        ggml_row_size(q->type, hparams.n_embd_head_k),
+                        ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
+                        0);
+                cb(q_nope, "q_nope", il);
 
-                struct ggml_tensor * Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, tmpq, n_embd_head, n_head,    n_tokens), inp_pos, nullptr,
+                // and {n_head * n_embd_head_qk_rope, n_tokens}
+                struct ggml_tensor * q_pe = ggml_view_3d(ctx0, q, n_embd_head_qk_rope, n_head, n_tokens,
+                        ggml_row_size(q->type, hparams.n_embd_head_k),
+                        ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
+                        ggml_row_size(q->type, n_embd_head_qk_nope));
+                cb(q_pe, "q_pe", il);
+
+                // {n_embd, kv_lora_rank + n_embd_head_qk_rope} * {n_embd, n_tokens} -> {kv_lora_rank + n_embd_head_qk_rope, n_tokens}
+                struct ggml_tensor * kv_pe_compresseed = ggml_mul_mat(ctx0, model.layers[il].wkv_a_mqa, cur);
+                cb(kv_pe_compresseed, "kv_pe_compresseed", il);
+
+                // split into {kv_lora_rank, n_tokens}
+                struct ggml_tensor * kv_compressed = ggml_view_2d(ctx0, kv_pe_compresseed, kv_lora_rank, n_tokens,
+                        kv_pe_compresseed->nb[1],
+                        0);
+                cb(kv_compressed, "kv_compressed", il);
+
+                // and {n_embd_head_qk_rope, n_tokens}
+                struct ggml_tensor * k_pe = ggml_view_3d(ctx0, kv_pe_compresseed, n_embd_head_qk_rope, 1, n_tokens,
+                        kv_pe_compresseed->nb[1],
+                        kv_pe_compresseed->nb[1],
+                        ggml_row_size(kv_pe_compresseed->type, kv_lora_rank));
+                cb(k_pe, "k_pe", il);
+
+                kv_compressed = ggml_cont(ctx0, kv_compressed); // TODO: the CUDA backend does not support non-contiguous norm
+                kv_compressed = llm_build_norm(ctx0, kv_compressed, hparams,
+                        model.layers[il].attn_kv_a_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(kv_compressed, "kv_compressed", il);
+
+                // {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)} * {kv_lora_rank, n_tokens} -> {n_head * (n_embd_head_qk_nope + n_embd_head_v), n_tokens}
+                struct ggml_tensor * kv = ggml_mul_mat(ctx0, model.layers[il].wkv_b, kv_compressed);
+                cb(kv, "kv", il);
+
+                // split into {n_head * n_embd_head_qk_nope, n_tokens}
+                struct ggml_tensor * k_nope = ggml_view_3d(ctx0, kv, n_embd_head_qk_nope, n_head, n_tokens,
+                        ggml_row_size(kv->type, n_embd_head_qk_nope + hparams.n_embd_head_v),
+                        ggml_row_size(kv->type, n_head * (n_embd_head_qk_nope + hparams.n_embd_head_v)),
+                        0);
+                cb(k_nope, "k_nope", il);
+
+                // and {n_head * n_embd_head_v, n_tokens}
+                struct ggml_tensor * v_states = ggml_view_3d(ctx0, kv, hparams.n_embd_head_v, n_head, n_tokens,
+                        ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)),
+                        ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)*n_head),
+                        ggml_row_size(kv->type, (n_embd_head_qk_nope)));
+                cb(v_states, "v_states", il);
+
+                v_states = ggml_cont(ctx0, v_states);
+                cb(v_states, "v_states", il);
+
+                v_states = ggml_view_2d(ctx0, v_states, hparams.n_embd_head_v * n_head, n_tokens,
+                    ggml_row_size(kv->type, hparams.n_embd_head_v * n_head),
+                    0);
+                cb(v_states, "v_states", il);
+
+                q_pe = ggml_cont(ctx0, q_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
+                q_pe = ggml_rope_ext(
+                    ctx0, q_pe, inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
+                    ext_factor, attn_factor_scaled, beta_fast, beta_slow
                 );
-                cb(Qcur, "Qcur", il);
+                cb(q_pe, "q_pe", il);
 
-                struct ggml_tensor * Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, tmpk, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                // shared RoPE key
+                k_pe = ggml_cont(ctx0, k_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
+                k_pe = ggml_rope_ext(
+                    ctx0, k_pe, inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
+                    ext_factor, attn_factor_scaled, beta_fast, beta_slow
                 );
-                cb(Kcur, "Kcur", il);
+                cb(k_pe, "k_pe", il);
+
+                struct ggml_tensor * q_states = ggml_concat(ctx0, q_nope, q_pe, 0);
+                cb(q_states, "q_states", il);
+
+                struct ggml_tensor * k_states = ggml_concat(ctx0, k_nope, ggml_repeat(ctx0, k_pe, q_pe), 0);
+                cb(k_states, "k_states", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+                        model.layers[il].wo, NULL,
+                        k_states, v_states, q_states, KQ_mask, n_tokens, kv_head, n_kv, kq_scale, cb, il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+                n_tokens = n_outputs;
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
             }
 
-            // add the input
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
             cb(ffn_inp, "ffn_inp", il);
 
-            // FF
-            {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm,
-                        model.layers[il].ffn_norm_b,
-                        LLM_NORM, cb, il);
-                cb(cur, "ffn_norm", il);
+            cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_norm", il);
 
+            if ((uint32_t) il < hparams.n_layer_dense_lead) {
                 cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        NULL,                      NULL,                        NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                        model.layers[il].ffn_up,   NULL, NULL,
+                        model.layers[il].ffn_gate, NULL, NULL,
+                        model.layers[il].ffn_down, NULL, NULL,
                         NULL,
-                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
                 cb(cur, "ffn_out", il);
+            } else {
+                // MoE branch
+                ggml_tensor * moe_out =
+                        llm_build_moe_ffn(ctx0, lctx, cur,
+                            model.layers[il].ffn_gate_inp,
+                            model.layers[il].ffn_up_exps,
+                            model.layers[il].ffn_gate_exps,
+                            model.layers[il].ffn_down_exps,
+                            model.layers[il].ffn_exp_probs_b,
+                            n_expert, n_expert_used,
+                            LLM_FFN_SILU, hparams.expert_weights_norm,
+                            true, hparams.expert_weights_scale,
+                            (enum llama_expert_gating_func_type) hparams.expert_gating_func,
+                            cb, il);
+                cb(moe_out, "ffn_moe_out", il);
+
+                // FFN shared expert
+                {
+                    ggml_tensor * ffn_shexp = llm_build_ffn(ctx0, lctx, cur,
+                            model.layers[il].ffn_up_shexp,   NULL, NULL,
+                            model.layers[il].ffn_gate_shexp, NULL, NULL,
+                            model.layers[il].ffn_down_shexp, NULL, NULL,
+                            NULL,
+                            LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                    cb(ffn_shexp, "ffn_shexp", il);
+
+                    cur = ggml_add(ctx0, moe_out, ffn_shexp);
+                    cb(cur, "ffn_out", il);
+                }
             }
 
             cur = ggml_add(ctx0, cur, ffn_inp);
@@ -13521,13 +8977,15 @@ struct llm_build_context {
             inpL = cur;
         }
 
-        cur = llm_build_norm(ctx0, inpL, hparams,
-                model.output_norm,
-                model.output_norm_b,
-                LLM_NORM, cb, -1);
+        cur = inpL;
+
+        cur = llm_build_norm(ctx0, cur, hparams,
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
         cb(cur, "result_norm", -1);
 
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        // lm_head
+        cur = ggml_mul_mat(ctx0, model.output, cur);
         cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
@@ -13535,12 +8993,11 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_orion() {
+    struct ggml_cgraph * build_bitnet() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
@@ -13556,38 +9013,48 @@ struct llm_build_context {
         for (int il = 0; il < n_layer; ++il) {
             struct ggml_tensor * inpSA = inpL;
 
-            // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, model.layers[il].attn_norm_b,
-                    LLM_NORM, cb, il);
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
             cb(cur, "attn_norm", il);
 
             // self-attention
             {
                 // compute Q and K and RoPE them
                 struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                if (model.layers[il].wq_scale) {
+                    Qcur = ggml_mul(ctx0, Qcur, model.layers[il].wq_scale);
+                }
                 cb(Qcur, "Qcur", il);
-                // if (model.layers[il].bq) {
-                //     Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                //     cb(Qcur, "Qcur", il);
-                // }
+                if (model.layers[il].bq) {
+                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
+                    cb(Qcur, "Qcur", il);
+                }
 
+                // B1.K
                 struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                if (model.layers[il].wk_scale) {
+                    Kcur = ggml_mul(ctx0, Kcur, model.layers[il].wk_scale);
+                }
                 cb(Kcur, "Kcur", il);
-                // if (model.layers[il].bk) {
-                //     Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                //     cb(Kcur, "Kcur", il);
-                // }
+                if (model.layers[il].bk) {
+                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
+                    cb(Kcur, "Kcur", il);
+                }
 
+                // B1.V
                 struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                if (model.layers[il].wv_scale) {
+                    Vcur = ggml_mul(ctx0, Vcur, model.layers[il].wv_scale);
+                }
                 cb(Vcur, "Vcur", il);
-                // if (model.layers[il].bv) {
-                //     Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                //     cb(Vcur, "Vcur", il);
-                // }
-
-                Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens), inp_pos, nullptr,
+                if (model.layers[il].bv) {
+                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
+                    cb(Vcur, "Vcur", il);
+                }
+
+                Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
@@ -13601,8 +9068,22 @@ struct llm_build_context {
                 cb(Kcur, "Kcur", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
+                        NULL, NULL,
                         Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+
+                cur = llm_build_norm(ctx0, cur, hparams,
+                        model.layers[il].attn_sub_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "attn_sub_norm", il);
+
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo, cur);
+                if (model.layers[il].wo_scale) {
+                    cur = ggml_mul(ctx0, cur, model.layers[il].wo_scale);
+                }
+                if (model.layers[il].bo) {
+                    cur = ggml_add(ctx0, cur, model.layers[il].bo);
+                }
+                cb(cur, "attn_o_out", il);
             }
 
             if (il == n_layer - 1) {
@@ -13615,22 +9096,32 @@ struct llm_build_context {
             struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
             cb(ffn_inp, "ffn_inp", il);
 
-            // feed-forward network
+            // feed-forward forward
             cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    model.layers[il].ffn_norm, model.layers[il].ffn_norm_b,
-                    LLM_NORM, cb, il);
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
             cb(cur, "ffn_norm", il);
 
             cur = llm_build_ffn(ctx0, lctx, cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
+                    model.layers[il].ffn_up,   NULL, model.layers[il].ffn_up_scale,
+                    model.layers[il].ffn_gate, NULL, model.layers[il].ffn_gate_scale,
+                    NULL,                      NULL, NULL,
                     NULL,
                     LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-            cb(cur, "ffn_out", il);
+            cb(cur, "ffn_sub_out", il);
+
+            cur = llm_build_norm(ctx0, cur, hparams,
+                            model.layers[il].ffn_sub_norm, NULL,
+                            LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_sub_norm", il);
+
+            cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].ffn_down, cur);
+            if (model.layers[il].ffn_down_scale) {
+                cur = ggml_mul(ctx0, cur, model.layers[il].ffn_down_scale);
+            }
+            cb(cur, "ffn_down", il);
 
             cur = ggml_add(ctx0, cur, ffn_inp);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
             // input for next layer
@@ -13640,92 +9131,99 @@ struct llm_build_context {
         cur = inpL;
 
         cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, model.output_norm_b,
-                LLM_NORM, cb, -1);
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
         cb(cur, "result_norm", -1);
 
         // lm_head
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        // FIXME: do not use model.tok_embd directly, duplicate as model.output
+        cur = llm_build_lora_mm(lctx, ctx0, model.tok_embd, cur);
         cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
-
         return gf;
     }
 
-    struct ggml_cgraph * build_internlm2() {
+    struct ggml_cgraph * build_t5_enc() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
+
         const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
 
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
+        GGML_ASSERT(lctx.is_encoding);
+        struct ggml_tensor * pos_bucket_enc = llm_build_pos_bucket(false);
 
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
+        struct ggml_tensor * KQ_mask_enc = build_inp_KQ_mask(false);
 
         for (int il = 0; il < n_layer; ++il) {
             struct ggml_tensor * inpSA = inpL;
 
             // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
+                    model.layers[il].attn_norm_enc, NULL,
                     LLM_NORM_RMS, cb, il);
             cb(cur, "attn_norm", il);
 
             // self-attention
             {
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq_enc, cur);
                 cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
 
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk_enc, cur);
                 cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
 
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv_enc, cur);
                 cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
 
-                Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur", il);
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
+                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
 
-                Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur", il);
+                struct ggml_tensor * q =                 ggml_permute(ctx0, Qcur, 0, 2, 1, 3);
+                struct ggml_tensor * k = ggml_cont(ctx0, ggml_permute(ctx0, Kcur, 0, 2, 1, 3));
 
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+                struct ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
+                cb(kq, "kq", il);
+
+                struct ggml_tensor * attn_rel_b = model.layers[il].attn_rel_b_enc ? model.layers[il].attn_rel_b_enc : model.layers[0].attn_rel_b_enc;
+                struct ggml_tensor * pos_bias = llm_build_pos_bias(pos_bucket_enc, attn_rel_b);
+                struct ggml_tensor * kq_b = ggml_add(ctx0, kq, pos_bias);
+                cb(kq_b, "kq_b", il);
+
+                kq = ggml_soft_max_ext(ctx0, kq_b, KQ_mask_enc, 1.0f, hparams.f_max_alibi_bias);
+                cb(kq, "kq_soft_max_ext", il);
+
+                struct ggml_tensor * v = ggml_cont(ctx0, ggml_transpose(ctx0, ggml_reshape_2d(ctx0, Vcur, n_embd_gqa, n_tokens)));
+                cb(v, "v", il);
+
+                struct ggml_tensor * kqv = ggml_mul_mat(ctx0, ggml_reshape_3d(ctx0, v, n_tokens, n_embd_head, n_head_kv), kq);
+                cb(kqv, "kqv", il);
+
+                struct ggml_tensor * kqv_merged = ggml_permute(ctx0, kqv, 0, 2, 1, 3);
+                cb(kqv_merged, "kqv_merged", il);
+
+                cur = ggml_cont_2d(ctx0, kqv_merged, n_embd_gqa, n_tokens);
+                cb(cur, "kqv_merged_cont", il);
+
+                ggml_build_forward_expand(gf, cur);
+
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo_enc, cur);
+                cb(cur, "kqv_out", il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                n_tokens = n_outputs;
                 cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
                 inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
             }
@@ -13734,21 +9232,31 @@ struct llm_build_context {
             cb(ffn_inp, "ffn_inp", il);
 
             // feed-forward network
-            cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm", il);
+            {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm_enc, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "ffn_norm", il);
 
-            cur = llm_build_ffn(ctx0, lctx, cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-            cb(cur, "ffn_out", il);
+                // T5 uses relu, flan-T5 uses gelu-gated
+                cur = llm_build_ffn(ctx0, lctx, cur,
+                        model.layers[il].ffn_up_enc,   NULL, NULL,
+                        model.layers[il].ffn_gate_enc, NULL, NULL,
+                        model.layers[il].ffn_down_enc, NULL, NULL,
+                        NULL,
+                        model.layers[il].ffn_gate_enc ? LLM_FFN_GELU : LLM_FFN_RELU,
+                        model.layers[il].ffn_gate_enc ? LLM_FFN_PAR  : LLM_FFN_SEQ,
+                        cb, il);
+                cb(cur, "ffn_out", il);
+            }
 
             cur = ggml_add(ctx0, cur, ffn_inp);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "ffn_out", il);
+
+            ggml_tensor * layer_dir = lctx.cvec.tensor_for(il);
+            if (layer_dir != nullptr) {
+                cur = ggml_add(ctx0, cur, layer_dir);
+            }
             cb(cur, "l_out", il);
 
             // input for next layer
@@ -13756,179 +9264,172 @@ struct llm_build_context {
         }
 
         cur = inpL;
+        cb(cur, "result_embd", -1);
 
         cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
+                model.output_norm_enc, NULL,
                 LLM_NORM_RMS, cb, -1);
         cb(cur, "result_norm", -1);
 
-        // lm_head
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
         ggml_build_forward_expand(gf, cur);
 
         return gf;
     }
 
-    struct ggml_cgraph * build_minicpm3() {
+    struct ggml_cgraph * build_t5_dec() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        //TODO: if the model varies, these parameters need to be read from the model
-        const int64_t n_embd_base = 256;
-        const float scale_embd  = 12.0f;
-        const float scale_depth = 1.4f;
-        const float kq_scale = 1.0f / sqrtf(float(hparams.n_embd_head_k));
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
 
-        const uint32_t n_embd_head_qk_rope = hparams.n_rot;
-        const uint32_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot;
-        const uint32_t kv_lora_rank = hparams.n_lora_kv;
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
 
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-        // scale the input embeddings
-        inpL = ggml_scale(ctx0, inpL, scale_embd);
-        cb(inpL, "inp_scaled", -1);
+        GGML_ASSERT(!lctx.is_encoding);
+        GGML_ASSERT(n_outputs_enc > 0 && "call llama_encode() first");
 
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
+        struct ggml_tensor * embd_enc       = llm_build_inp_embd_enc();
+        struct ggml_tensor * pos_bucket_dec = llm_build_pos_bucket(true);
 
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
+        struct ggml_tensor * KQ_mask_dec   = build_inp_KQ_mask();
+        struct ggml_tensor * KQ_mask_cross = llm_build_inp_KQ_mask_cross();
 
         for (int il = 0; il < n_layer; ++il) {
             struct ggml_tensor * inpSA = inpL;
 
-            struct ggml_tensor * rope_factors = build_rope_factors(il);
             // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
                     model.layers[il].attn_norm, NULL,
                     LLM_NORM_RMS, cb, il);
             cb(cur, "attn_norm", il);
 
-            // self_attention
+            // self-attention
             {
-                struct ggml_tensor * q = NULL;
-                // {n_embd, q_lora_rank} * {n_embd, n_tokens} -> {q_lora_rank, n_tokens}
-                q = ggml_mul_mat(ctx0, model.layers[il].wq_a, cur);
-                cb(q, "q", il);
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
+                cb(Qcur, "Qcur", il);
 
-                q = llm_build_norm(ctx0, q, hparams,
-                        model.layers[il].attn_q_a_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(q, "q", il);
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
+                cb(Kcur, "Kcur", il);
 
-                // {q_lora_rank, n_head * hparams.n_embd_head_k} * {q_lora_rank, n_tokens} -> {n_head * hparams.n_embd_head_k, n_tokens}
-                q = ggml_mul_mat(ctx0, model.layers[il].wq_b, q);
-                cb(q, "q", il);
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
+                cb(Vcur, "Vcur", il);
 
-                // split into {n_head * n_embd_head_qk_nope, n_tokens}
-                struct ggml_tensor * q_nope = ggml_view_3d(ctx0, q, n_embd_head_qk_nope, n_head, n_tokens,
-                        ggml_row_size(q->type, hparams.n_embd_head_k),
-                        ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
-                        0);
-                cb(q_nope, "q_nope", il);
+                llm_build_kv_store(ctx0, hparams, cparams, kv_self, gf, Kcur, Vcur, n_tokens, kv_head, cb, il);
 
-                // and {n_head * n_embd_head_qk_rope, n_tokens}
-                struct ggml_tensor * q_pe = ggml_view_3d(ctx0, q, n_embd_head_qk_rope, n_head, n_tokens,
-                        ggml_row_size(q->type, hparams.n_embd_head_k),
-                        ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
-                        ggml_row_size(q->type, n_embd_head_qk_nope));
-                cb(q_pe, "q_pe", il);
+                struct ggml_tensor * k =
+                    ggml_view_3d(ctx0, kv_self.k_l[il],
+                            n_embd_head_k, n_kv, n_head_kv,
+                            ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa),
+                            ggml_row_size(kv_self.k_l[il]->type, n_embd_head_k),
+                            0);
+                cb(k, "k", il);
 
-                // {n_embd, kv_lora_rank + n_embd_head_qk_rope} * {n_embd, n_tokens} -> {kv_lora_rank + n_embd_head_qk_rope, n_tokens}
-                struct ggml_tensor * kv_pe_compresseed = ggml_mul_mat(ctx0, model.layers[il].wkv_a_mqa, cur);
-                cb(kv_pe_compresseed, "kv_pe_compresseed", il);
+                struct ggml_tensor * v =
+                    ggml_view_3d(ctx0, kv_self.v_l[il],
+                            n_kv, n_embd_head_v, n_head_kv,
+                            ggml_element_size(kv_self.v_l[il])*n_ctx,
+                            ggml_element_size(kv_self.v_l[il])*n_ctx*n_embd_head_v,
+                            0);
+                cb(v, "v", il);
 
-                // split into {kv_lora_rank, n_tokens}
-                struct ggml_tensor * kv_compressed = ggml_view_2d(ctx0, kv_pe_compresseed, kv_lora_rank, n_tokens,
-                        kv_pe_compresseed->nb[1],
-                        0);
-                cb(kv_compressed, "kv_compressed", il);
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
 
-                // and {n_embd_head_qk_rope, n_tokens}
-                struct ggml_tensor * k_pe = ggml_view_3d(ctx0, kv_pe_compresseed, n_embd_head_qk_rope, 1, n_tokens,
-                        kv_pe_compresseed->nb[1],
-                        kv_pe_compresseed->nb[1],
-                        ggml_row_size(kv_pe_compresseed->type, kv_lora_rank));
-                cb(k_pe, "k_pe", il);
+                struct ggml_tensor * q = ggml_permute(ctx0, Qcur, 0, 2, 1, 3);
 
-                kv_compressed = ggml_cont(ctx0, kv_compressed); // TODO: the CUDA backend does not support non-contiguous norm
-                kv_compressed = llm_build_norm(ctx0, kv_compressed, hparams,
-                        model.layers[il].attn_kv_a_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(kv_compressed, "kv_compressed", il);
+                struct ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
+                cb(kq, "kq", il);
 
-                // {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)} * {kv_lora_rank, n_tokens} -> {n_head * (n_embd_head_qk_nope + n_embd_head_v), n_tokens}
-                struct ggml_tensor * kv = ggml_mul_mat(ctx0, model.layers[il].wkv_b, kv_compressed);
-                cb(kv, "kv", il);
+                struct ggml_tensor * attn_rel_b = model.layers[il].attn_rel_b ? model.layers[il].attn_rel_b : model.layers[0].attn_rel_b;
+                struct ggml_tensor * pos_bias = llm_build_pos_bias(pos_bucket_dec, attn_rel_b);
+                struct ggml_tensor * kq_b = ggml_add(ctx0, kq, pos_bias);
+                cb(kq_b, "kq_b", il);
 
-                // split into {n_head * n_embd_head_qk_nope, n_tokens}
-                struct ggml_tensor * k_nope = ggml_view_3d(ctx0, kv, n_embd_head_qk_nope, n_head, n_tokens,
-                        ggml_row_size(kv->type, n_embd_head_qk_nope + hparams.n_embd_head_v),
-                        ggml_row_size(kv->type, n_head * (n_embd_head_qk_nope + hparams.n_embd_head_v)),
-                        0);
-                cb(k_nope, "k_nope", il);
+                kq = ggml_soft_max_ext(ctx0, kq_b, KQ_mask_dec, 1.0f, hparams.f_max_alibi_bias);
+                cb(kq, "kq_soft_max_ext", il);
 
-                // and {n_head * n_embd_head_v, n_tokens}
-                struct ggml_tensor * v_states = ggml_view_3d(ctx0, kv, hparams.n_embd_head_v, n_head, n_tokens,
-                        ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)),
-                        ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)*n_head),
-                        ggml_row_size(kv->type, (n_embd_head_qk_nope)));
-                cb(v_states, "v_states", il);
+                struct ggml_tensor * kqv = ggml_mul_mat(ctx0, v, kq);
+                cb(kqv, "kqv", il);
 
-                v_states = ggml_cont(ctx0, v_states);
-                cb(v_states, "v_states", il);
+                struct ggml_tensor * kqv_merged = ggml_permute(ctx0, kqv, 0, 2, 1, 3);
+                cb(kqv_merged, "kqv_merged", il);
 
-                v_states = ggml_view_2d(ctx0, v_states, hparams.n_embd_head_v * n_head, n_tokens,
-                    ggml_row_size(kv->type, hparams.n_embd_head_v * n_head),
-                    0);
-                cb(v_states, "v_states", il);
+                cur = ggml_cont_2d(ctx0, kqv_merged, n_embd_gqa, n_tokens);
+                cb(cur, "kqv_merged_cont", il);
 
-                q_pe = ggml_cont(ctx0, q_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
-                q_pe = ggml_rope_ext(
-                    ctx0, q_pe, inp_pos, rope_factors,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(q_pe, "q_pe", il);
+                ggml_build_forward_expand(gf, cur);
 
-                // shared RoPE key
-                k_pe = ggml_cont(ctx0, k_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
-                k_pe = ggml_rope_ext(
-                    ctx0, k_pe, inp_pos, rope_factors,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(k_pe, "k_pe", il);
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo, cur);
+                cb(cur, "kqv_out", il);
+            }
 
-                struct ggml_tensor * q_states = ggml_concat(ctx0, q_nope, q_pe, 0);
-                cb(q_states, "q_states", il);
+            cur = ggml_add(ctx0, cur, inpSA);
+            cb(cur, "cross_inp", il);
 
-                struct ggml_tensor * k_states = ggml_concat(ctx0, k_nope, ggml_repeat(ctx0, k_pe, q_pe), 0);
-                cb(k_states, "k_states", il);
+            struct ggml_tensor * inpCA = cur;
 
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
-                        k_states, v_states, q_states, KQ_mask, n_tokens, kv_head, n_kv, kq_scale, cb, il);
+            // norm
+            cur = llm_build_norm(ctx0, cur, hparams,
+                    model.layers[il].attn_norm_cross, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "attn_norm_cross", il);
+
+            // cross-attention
+            {
+                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq_cross, cur);
+                cb(Qcur, "Qcur", il);
+
+                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk_cross, embd_enc);
+                cb(Kcur, "Kcur", il);
+
+                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv_cross, embd_enc);
+                cb(Vcur, "Vcur", il);
+
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
+                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_outputs_enc);
+
+                struct ggml_tensor * q =                 ggml_permute(ctx0, Qcur, 0, 2, 1, 3);
+                struct ggml_tensor * k = ggml_cont(ctx0, ggml_permute(ctx0, Kcur, 0, 2, 1, 3));
+
+                struct ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
+                cb(kq, "kq", il);
+
+                kq = ggml_soft_max_ext(ctx0, kq, KQ_mask_cross, 1.0f, hparams.f_max_alibi_bias);
+                cb(kq, "kq_soft_max_ext", il);
+
+                struct ggml_tensor * v = ggml_cont(ctx0, ggml_transpose(ctx0, ggml_reshape_2d(ctx0, Vcur, n_embd_gqa, n_outputs_enc)));
+                cb(v, "v", il);
+
+                struct ggml_tensor * kqv = ggml_mul_mat(ctx0, ggml_reshape_3d(ctx0, v, n_outputs_enc, n_embd_head, n_head_kv), kq);
+                cb(kqv, "kqv", il);
+
+                struct ggml_tensor * kqv_merged = ggml_permute(ctx0, kqv, 0, 2, 1, 3);
+                cb(kqv_merged, "kqv_merged", il);
+
+                cur = ggml_cont_2d(ctx0, kqv_merged, n_embd_gqa, n_tokens);
+                cb(cur, "kqv_merged_cont", il);
+
+                ggml_build_forward_expand(gf, cur);
+
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo_cross, cur);
+                cb(cur, "kqv_out", il);
             }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+                n_tokens = n_outputs;
                 cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
                 inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
+                inpCA = ggml_get_rows(ctx0, inpCA, inp_out_ids);
             }
 
-            // scale_res - scale the hidden states for residual connection
-            const float scale_res = scale_depth/sqrtf(float(n_layer));
-            cur = ggml_scale(ctx0, cur, scale_res);
-            cb(cur, "hidden_scaled", il);
-
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpCA);
             cb(ffn_inp, "ffn_inp", il);
 
             // feed-forward network
@@ -13938,21 +9439,25 @@ struct llm_build_context {
                         LLM_NORM_RMS, cb, il);
                 cb(cur, "ffn_norm", il);
 
+                // T5 uses relu, flan-T5 uses gelu-gated
                 cur = llm_build_ffn(ctx0, lctx, cur,
                         model.layers[il].ffn_up,   NULL, NULL,
                         model.layers[il].ffn_gate, NULL, NULL,
                         model.layers[il].ffn_down, NULL, NULL,
                         NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+                        model.layers[il].ffn_gate_enc ? LLM_FFN_GELU : LLM_FFN_RELU,
+                        model.layers[il].ffn_gate_enc ? LLM_FFN_PAR : LLM_FFN_SEQ,
+                        cb, il);
                 cb(cur, "ffn_out", il);
             }
 
-            // scale the hidden states for residual connection
-            cur = ggml_scale(ctx0, cur, scale_res);
-            cb(cur, "hidden_scaled_ffn", il);
-
             cur = ggml_add(ctx0, cur, ffn_inp);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "ffn_out", il);
+
+            ggml_tensor * layer_dir = lctx.cvec.tensor_for(il);
+            if (layer_dir != nullptr) {
+                cur = ggml_add(ctx0, cur, layer_dir);
+            }
             cb(cur, "l_out", il);
 
             // input for next layer
@@ -13960,17 +9465,13 @@ struct llm_build_context {
         }
 
         cur = inpL;
+        cb(cur, "result_embd", -1);
 
         cur = llm_build_norm(ctx0, cur, hparams,
                 model.output_norm, NULL,
                 LLM_NORM_RMS, cb, -1);
         cb(cur, "result_norm", -1);
 
-        // lm_head scaling
-        const float scale_lmhead = float(n_embd_base)/float(n_embd);
-        cur = ggml_scale(ctx0, cur, scale_lmhead);
-        cb(cur, "lmhead_scaling", -1);
-
         // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
         cb(cur, "result_output", -1);
@@ -13980,62 +9481,49 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_gemma() {
+    struct ggml_cgraph * build_jais() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        const int64_t n_embd_head_k = hparams.n_embd_head_k;
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
 
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-        inpL = ggml_scale(ctx0, inpL, sqrtf(n_embd));
-        cb(inpL, "inp_scaled", -1);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
         for (int il = 0; il < n_layer; ++il) {
-            // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
+                    model.layers[il].attn_norm,
+                    model.layers[il].attn_norm_b,
+                    LLM_NORM, cb, il);
             cb(cur, "attn_norm", il);
 
             // self-attention
             {
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
+                cb(cur, "bqkv", il);
 
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
+                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*cur->nb[0]*(n_embd)));
+                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*cur->nb[0]*(n_embd)));
+                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*cur->nb[0]*(n_embd + n_embd_gqa)));
 
-                Qcur = ggml_rope_ext(
-                        ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head_k, n_head,    n_tokens), inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
                 cb(Qcur, "Qcur", il);
-
-                Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd_head_k)));
-                cb(Qcur, "Qcur_scaled", il);
-
-                Kcur = ggml_rope_ext(
-                        ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head_k, n_head_kv, n_tokens), inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
                 cb(Kcur, "Kcur", il);
+                cb(Vcur, "Vcur", il);
+
+                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f, cb, il);
+                        model.layers[il].wo, model.layers[il].bo,
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/float(n_embd_head), cb, il);
             }
 
             if (il == n_layer - 1) {
@@ -14045,42 +9533,39 @@ struct llm_build_context {
                 inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
             }
 
-            struct ggml_tensor * sa_out = ggml_add(ctx0, cur, inpL);
-            cb(sa_out, "sa_out", il);
-
-            cur = llm_build_norm(ctx0, sa_out, hparams,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm", il);
+            // add the input
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
+            cb(ffn_inp, "ffn_inp", il);
 
-            // feed-forward network
+            // FF
             {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm,
+                        model.layers[il].ffn_norm_b,
+                        LLM_NORM, cb, il);
+                cb(cur, "ffn_norm", il);
+
                 cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
+                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                        model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
+                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
                         NULL,
-                        LLM_FFN_GELU, LLM_FFN_PAR, cb, il);
+                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
                 cb(cur, "ffn_out", il);
             }
 
-            cur = ggml_add(ctx0, cur, sa_out);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
+            inpL = ggml_add(ctx0, cur, ffn_inp);
+            cb(inpL, "l_out", il);
         }
 
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
+        cur = llm_build_norm(ctx0, inpL, hparams,
+                model.output_norm,
+                model.output_norm_b,
+                LLM_NORM, cb, -1);
         cb(cur, "result_norm", -1);
 
-        // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+
         cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
@@ -14088,134 +9573,113 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_gemma2() {
+    struct ggml_cgraph * build_chatglm() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        const int64_t n_embd_head_k = hparams.n_embd_head_k;
+        const int64_t n_embd_head = hparams.n_embd_head_v;
+        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
+        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
 
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-        inpL = ggml_scale(ctx0, inpL, sqrtf(n_embd));
-        cb(inpL, "inp_scaled", -1);
-
         // inp_pos - contains the positions
         struct ggml_tensor * inp_pos = build_inp_pos();
 
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        // gemma 2 requires different mask for layers using sliding window (SWA)
-        struct ggml_tensor * KQ_mask     = build_inp_KQ_mask(true);
-        struct ggml_tensor * KQ_mask_swa = build_inp_KQ_mask_swa(true);
+        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
         for (int il = 0; il < n_layer; ++il) {
-            // (il % 2) layers use SWA
-            struct ggml_tensor * KQ_mask_l = (il % 2 == 0) ? KQ_mask_swa : KQ_mask;
+            struct ggml_tensor * inpSA = inpL;
 
-            // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
+                    model.layers[il].attn_norm,
+                    NULL,
                     LLM_NORM_RMS, cb, il);
             cb(cur, "attn_norm", il);
 
             // self-attention
             {
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
+                struct ggml_tensor * Qcur = nullptr;
+                struct ggml_tensor * Kcur = nullptr;
+                struct ggml_tensor * Vcur = nullptr;
 
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
+                cb(cur, "wqkv", il);
 
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
+                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
+                cb(cur, "bqkv", il);
 
-                Qcur = ggml_rope_ext(
-                        ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head_k, n_head,    n_tokens), inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
-                cb(Qcur, "Qcur", il);
+                Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
+                Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
+                Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
 
-                // ref: https://github.com/google/gemma_pytorch/commit/03e657582d17cb5a8617ebf333c1c16f3694670e
-                switch (model.type) {
-                    case e_model::MODEL_2B:
-                    case e_model::MODEL_9B:  Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd_head_k)));   break;
-                    case e_model::MODEL_27B: Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd / n_head))); break;
-                    default: GGML_ABORT("fatal error");
-                };
-                cb(Qcur, "Qcur_scaled", il);
+                cb(Qcur, "Qcur", il);
+                cb(Kcur, "Kcur", il);
+                cb(Vcur, "Vcur", il);
+                //printf("freq_base: %f freq_scale: %f ext_factor: %f attn_factor: %f\n", freq_base, freq_scale, ext_factor, attn_factor);
+                Qcur = ggml_rope_ext(
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Qcur, "Qcur_rope", il);
 
                 Kcur = ggml_rope_ext(
-                        ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head_k, n_head_kv, n_tokens), inp_pos, nullptr,
-                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                        ext_factor, attn_factor, beta_fast, beta_slow);
-                cb(Kcur, "Kcur", il);
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                    ext_factor, attn_factor, beta_fast, beta_slow
+                );
+                cb(Kcur, "Kcur_rope", il);
 
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
                         model.layers[il].wo, NULL,
-                        Kcur, Vcur, Qcur, KQ_mask_l, n_tokens, kv_head, n_kv, 1.0f, cb, il);
-            }
+                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
 
-            cur = llm_build_norm(ctx0, cur, hparams,
-                    model.layers[il].attn_post_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_post_norm", il);
+            }
 
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
             }
 
-            struct ggml_tensor * sa_out = ggml_add(ctx0, cur, inpL);
-            cb(sa_out, "sa_out", il);
-
-            cur = llm_build_norm(ctx0, sa_out, hparams,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm", il);
+            // Add the input
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
-            // feed-forward network
+            // FF
             {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm,
+                        NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "ffn_norm", il);
+
                 cur = llm_build_ffn(ctx0, lctx, cur,
                         model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
+                        NULL,                      NULL, NULL,
                         model.layers[il].ffn_down, NULL, NULL,
                         NULL,
-                        LLM_FFN_GELU, LLM_FFN_PAR, cb, il);
+                        LLM_FFN_SWIGLU, LLM_FFN_SEQ, cb, il);
                 cb(cur, "ffn_out", il);
-            }
-
-            cur = llm_build_norm(ctx0, cur, hparams,
-                model.layers[il].ffn_post_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-            cb(cur, "ffn_post_norm", -1);
 
-            cur = ggml_add(ctx0, cur, sa_out);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
+            }
 
-            // input for next layer
-            inpL = cur;
+            inpL = ggml_add(ctx0, cur, ffn_inp);
+            cb(inpL, "l_out", il);
         }
 
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
+        cur = llm_build_norm(ctx0, inpL, hparams,
+                model.output_norm,
+                NULL,
                 LLM_NORM_RMS, cb, -1);
         cb(cur, "result_norm", -1);
 
-        // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-
-        // final logit soft-capping
-        cur = ggml_scale(ctx0, cur, 1.0f / hparams.f_final_logit_softcapping);
-        cur = ggml_tanh(ctx0, cur);
-        cur = ggml_scale(ctx0, cur, hparams.f_final_logit_softcapping);
-
         cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
@@ -14223,13 +9687,12 @@ struct llm_build_context {
         return gf;
     }
 
-
-    struct ggml_cgraph * build_starcoder2() {
+    struct ggml_cgraph * build_nemotron() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
+        //GGML_ASSERT(n_embd_head == hparams.n_rot);
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
@@ -14247,7 +9710,8 @@ struct llm_build_context {
 
             // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, model.layers[il].attn_norm_b,
+                    model.layers[il].attn_norm,
+                    model.layers[il].attn_norm_b,
                     LLM_NORM, cb, il);
             cb(cur, "attn_norm", il);
 
@@ -14305,21 +9769,22 @@ struct llm_build_context {
             cb(ffn_inp, "ffn_inp", il);
 
             // feed-forward network
-
             cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    model.layers[il].ffn_norm, model.layers[il].ffn_norm_b,
+                    model.layers[il].ffn_norm,
+                    model.layers[il].ffn_norm_b,
                     LLM_NORM, cb, il);
             cb(cur, "ffn_norm", il);
 
             cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        NULL,                      NULL,                        NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                        NULL,
-                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
-            cb(cur, "ffn_out", il);
+                    model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
+                    NULL,                      NULL,                        NULL,
+                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
+                    NULL,
+                    LLM_FFN_RELU_SQR, LLM_FFN_SEQ, cb, il);
 
             cur = ggml_add(ctx0, cur, ffn_inp);
+            cb(cur, "ffn_out", il);
+
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
@@ -14343,67 +9808,15 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_mamba() {
+    struct ggml_cgraph * build_exaone() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        // {n_embd, n_tokens}
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        struct ggml_tensor * state_copy = build_inp_s_copy();
-        struct ggml_tensor * state_mask = build_inp_s_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
-
-            cur = llm_build_mamba(ctx0, lctx, ubatch, gf, cur,
-                    state_copy, state_mask,
-                    kv_head, n_kv, cb, il);
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
-            }
-
-            // residual
-            cur = ggml_add(ctx0, cur, inpL);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        // final rmsnorm
-        cur = llm_build_norm(ctx0, inpL, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        // lm_head
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    struct ggml_cgraph * build_command_r() {
-
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
+        // mutable variable, needed during the last layer of the computation to skip unused tokens
+        int32_t n_tokens = this->n_tokens;
 
         const int64_t n_embd_head = hparams.n_embd_head_v;
         GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        const float f_logit_scale = hparams.f_logit_scale;
+        GGML_ASSERT(n_embd_head == hparams.n_rot);
 
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
@@ -14417,16 +9830,19 @@ struct llm_build_context {
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
 
         for (int il = 0; il < n_layer; ++il) {
+            struct ggml_tensor * inpSA = inpL;
 
             // norm
             cur = llm_build_norm(ctx0, inpL, hparams,
                     model.layers[il].attn_norm, NULL,
-                    LLM_NORM, cb, il);
+                    LLM_NORM_RMS, cb, il);
             cb(cur, "attn_norm", il);
-            struct ggml_tensor * ffn_inp = cur;
 
             // self-attention
             {
+                // rope freq factors for llama3; may return nullptr for llama2 and other models
+                struct ggml_tensor * rope_factors = build_rope_factors(il);
+
                 // compute Q and K and RoPE them
                 struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
                 cb(Qcur, "Qcur", il);
@@ -14449,40 +9865,15 @@ struct llm_build_context {
                     cb(Vcur, "Vcur", il);
                 }
 
-                if (model.layers[il].attn_q_norm) {
-                    Qcur = ggml_view_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens,
-                                ggml_element_size(Qcur) * n_embd_head,
-                                ggml_element_size(Qcur) * n_embd_head * n_head,
-                                0);
-                    cb(Qcur, "Qcur", il);
-                    Kcur = ggml_view_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens,
-                                ggml_element_size(Kcur) * n_embd_head,
-                                ggml_element_size(Kcur) * n_embd_head * n_head_kv,
-                                0);
-                    cb(Kcur, "Kcur", il);
-
-                    Qcur = llm_build_norm(ctx0, Qcur, hparams,
-                                model.layers[il].attn_q_norm,
-                                NULL,
-                                LLM_NORM, cb, il);
-                    cb(Qcur, "Qcur", il);
-
-                    Kcur = llm_build_norm(ctx0, Kcur, hparams,
-                            model.layers[il].attn_k_norm,
-                            NULL,
-                            LLM_NORM, cb, il);
-                    cb(Kcur, "Kcur", il);
-                }
-
                 Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
+                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, rope_factors,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
                 cb(Qcur, "Qcur", il);
 
                 Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
+                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, rope_factors,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                     ext_factor, attn_factor, beta_fast, beta_slow
                 );
@@ -14496,27 +9887,31 @@ struct llm_build_context {
             if (il == n_layer - 1) {
                 // skip computing output for unused tokens
                 struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur     = ggml_get_rows(ctx0,     cur, inp_out_ids);
-                inpL    = ggml_get_rows(ctx0,    inpL, inp_out_ids);
-                ffn_inp = ggml_get_rows(ctx0, ffn_inp, inp_out_ids);
+                n_tokens = n_outputs;
+                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
+                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
             }
 
-            struct ggml_tensor * attn_out = cur;
+            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
+            cb(ffn_inp, "ffn_inp", il);
 
             // feed-forward network
-            {
-                cur = llm_build_ffn(ctx0, lctx, ffn_inp,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-                cb(cur, "ffn_out", il);
-            }
+            cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                    model.layers[il].ffn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+            cb(cur, "ffn_norm", il);
+
+            cur = llm_build_ffn(ctx0, lctx, cur,
+                    model.layers[il].ffn_up,   NULL, NULL,
+                    model.layers[il].ffn_gate, NULL, NULL,
+                    model.layers[il].ffn_down, NULL, NULL,
+                    NULL,
+                    LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
+            cb(cur, "ffn_out", il);
+
+            cur = ggml_add(ctx0, cur, ffn_inp);
+            cb(cur, "ffn_out", il);
 
-            // add together residual + FFN + self-attention
-            cur = ggml_add(ctx0, cur, inpL);
-            cur = ggml_add(ctx0, cur, attn_out);
             cur = lctx.cvec.apply_to(ctx0, cur, il);
             cb(cur, "l_out", il);
 
@@ -14528,31 +9923,138 @@ struct llm_build_context {
 
         cur = llm_build_norm(ctx0, cur, hparams,
                 model.output_norm, NULL,
-                LLM_NORM, cb, -1);
+                LLM_NORM_RMS, cb, -1);
         cb(cur, "result_norm", -1);
 
         // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output", -1);
 
-        if (f_logit_scale) {
-            cur = ggml_scale(ctx0, cur, f_logit_scale);
+        ggml_build_forward_expand(gf, cur);
+
+        return gf;
+    }
+
+    ggml_cgraph * build_rwkv6() {
+        ggml_cgraph *gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
+
+        // Token shift state dimensions should be 2 * n_emb
+        GGML_ASSERT(n_embd == hparams.n_embd_k_s() / 2);
+
+        const int64_t n_seqs = ubatch.n_seqs;
+        const int64_t n_seq_tokens = ubatch.n_seq_tokens;
+        const int64_t n_tokens = ubatch.n_tokens;
+        GGML_ASSERT(n_seqs != 0);
+        GGML_ASSERT(ubatch.equal_seqs);
+        GGML_ASSERT(n_tokens == n_seq_tokens * n_seqs);
+
+        struct ggml_tensor * cur;
+        struct ggml_tensor * inpL;
+        struct ggml_tensor * state_copy = build_inp_s_copy();
+        struct ggml_tensor * state_mask = build_inp_s_mask();
+
+        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
+        inpL = llm_build_norm(ctx0, inpL, hparams, model.tok_norm, model.tok_norm_b, LLM_NORM, cb, -1);
+
+        for (int il = 0; il < n_layer; ++il) {
+            const llama_layer * layer = &model.layers[il];
+
+            // (ab)using the KV cache to store the states
+            struct ggml_tensor * token_shift = llm_build_copy_mask_state(ctx0,
+                    gf, kv_self.k_l[il], state_copy, state_mask,
+                    hparams.n_embd_k_s(), kv_self.size, kv_head, n_kv, n_seqs);
+            struct ggml_tensor * wkv_states = llm_build_copy_mask_state(ctx0,
+                    gf, kv_self.v_l[il], state_copy, state_mask,
+                    hparams.n_embd_v_s(), kv_self.size, kv_head, n_kv, n_seqs);
+
+            cur = ggml_reshape_3d(ctx0, inpL, n_embd, n_seq_tokens, n_seqs);
+            token_shift = ggml_reshape_3d(ctx0, token_shift, n_embd, 2, n_seqs);
+
+            struct ggml_tensor * att_shift = ggml_view_3d(ctx0, token_shift, n_embd, 1, n_seqs, token_shift->nb[1], token_shift->nb[2], 0);
+            struct ggml_tensor * ffn_shift = ggml_view_3d(ctx0, token_shift, n_embd, 1, n_seqs, token_shift->nb[1], token_shift->nb[2], n_embd * ggml_element_size(token_shift));
+
+            struct ggml_tensor * x_norm_att = llm_build_norm(ctx0, cur, hparams, layer->attn_norm, layer->attn_norm_b, LLM_NORM, cb, il);
+            struct ggml_tensor * x_prev = ggml_concat(
+                ctx0,
+                att_shift,
+                ggml_view_3d(ctx0, x_norm_att, n_embd, n_seq_tokens - 1, n_seqs, x_norm_att->nb[1], x_norm_att->nb[2], 0),
+                1
+            );
+
+            cur = ggml_add(ctx0, cur, llm_build_rwkv6_time_mix(lctx, ctx0, layer, x_norm_att, x_prev, &wkv_states));
+            ggml_build_forward_expand(gf, cur);
+            ggml_build_forward_expand(
+                gf,
+                ggml_cpy(
+                    ctx0,
+                    wkv_states,
+                    ggml_view_1d(
+                        ctx0,
+                        kv_self.v_l[il],
+                        hparams.n_embd_v_s() * n_seqs,
+                        hparams.n_embd_v_s() * kv_head * ggml_element_size(kv_self.v_l[il])
+                    )
+                )
+            );
+
+            struct ggml_tensor * x_norm_ffn = llm_build_norm(ctx0, cur, hparams, layer->attn_norm_2, layer->attn_norm_2_b, LLM_NORM, cb, il);
+            x_prev = ggml_concat(
+                ctx0,
+                ffn_shift,
+                ggml_view_3d(ctx0, x_norm_ffn, n_embd, n_seq_tokens - 1, n_seqs, x_norm_ffn->nb[1], x_norm_ffn->nb[2], 0),
+                1
+            );
+            cur = ggml_add(ctx0, cur, llm_build_rwkv6_channel_mix(lctx, ctx0, layer, x_norm_ffn, x_prev));
+            ggml_build_forward_expand(gf, cur);
+
+            struct ggml_tensor * last_norm_att = ggml_view_3d(ctx0, x_norm_att, n_embd, 1, n_seqs, x_norm_att->nb[1], x_norm_att->nb[2], (n_seq_tokens-1)*n_embd*ggml_element_size(x_norm_att));
+            struct ggml_tensor * last_norm_ffn = ggml_view_3d(ctx0, x_norm_ffn, n_embd, 1, n_seqs, x_norm_ffn->nb[1], x_norm_ffn->nb[2], (n_seq_tokens-1)*n_embd*ggml_element_size(x_norm_ffn));
+
+            token_shift = ggml_concat(ctx0, last_norm_att, last_norm_ffn, 1);
+
+            ggml_build_forward_expand(
+                gf,
+                ggml_cpy(
+                    ctx0,
+                    ggml_view_1d(ctx0, token_shift, n_embd * n_seqs * 2, 0),
+                    ggml_view_1d(ctx0, kv_self.k_l[il], hparams.n_embd_k_s() * n_seqs, hparams.n_embd_k_s() * kv_head * ggml_element_size(kv_self.k_l[il]))
+                )
+            );
+
+            if (hparams.rescale_every_n_layers != 0 && (il + 1) % hparams.rescale_every_n_layers == 0) {
+                cur = ggml_scale(ctx0, cur, 0.5F);
+            }
+
+            cur = lctx.cvec.apply_to(ctx0, cur, il);
+            cb(cur, "l_out", il);
+
+            // input for next layer
+            inpL = cur;
         }
 
+        cur = inpL;
+        struct ggml_tensor * inp_out_ids = build_inp_out_ids();
+        cur = ggml_reshape_2d(ctx0, cur, n_embd, n_tokens);
+        cur = ggml_get_rows(ctx0, cur, inp_out_ids);
+
+        cur = llm_build_norm(ctx0, cur, hparams, model.output_norm, model.output_norm_b, LLM_NORM, cb, -1);
+        cb(cur, "result_norm", -1);
+
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
         cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
 
         return gf;
-
     }
 
-    // ref: https://allenai.org/olmo
+    // ref: https://github.com/facebookresearch/chameleon
     // based on the original build_llama() function, changes:
-    //   * non-parametric layer norm
-    //   * clamp qkv
+    //   * qk-norm
+    //   * swin-norm
     //   * removed bias
     //   * removed MoE
-    struct ggml_cgraph * build_olmo() {
+    struct ggml_cgraph * build_chameleon() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
         // mutable variable, needed during the last layer of the computation to skip unused tokens
@@ -14577,33 +10079,53 @@ struct llm_build_context {
             struct ggml_tensor * inpSA = inpL;
 
             // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    NULL, NULL,
-                    LLM_NORM, cb, il);
-            cb(cur, "attn_norm", il);
+            if (hparams.swin_norm) {
+                cur = inpL;
+            } else {
+                cur = llm_build_norm(ctx0, inpL, hparams,
+                    model.layers[il].attn_norm, NULL,
+                    LLM_NORM_RMS, cb, il);
+                cb(cur, "attn_norm", il);
+            }
 
             // self-attention
             {
                 // compute Q and K and RoPE them
                 struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
                 cb(Qcur, "Qcur", il);
-                if (hparams.f_clamp_kqv > 0.0f) {
-                    Qcur = ggml_clamp(ctx0, Qcur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
-                    cb(Qcur, "Qcur", il);
-                }
 
                 struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
                 cb(Kcur, "Kcur", il);
-                if (hparams.f_clamp_kqv > 0.0f) {
-                    Kcur = ggml_clamp(ctx0, Kcur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
-                    cb(Kcur, "Kcur", il);
-                }
 
                 struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
                 cb(Vcur, "Vcur", il);
-                if (hparams.f_clamp_kqv > 0.0f) {
-                    Vcur = ggml_clamp(ctx0, Vcur, -hparams.f_clamp_kqv, hparams.f_clamp_kqv);
-                    cb(Vcur, "Vcur", il);
+
+                if (model.layers[il].attn_q_norm) {
+                    Qcur = ggml_view_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens,
+                                ggml_element_size(Qcur) * n_embd_head,
+                                ggml_element_size(Qcur) * n_embd_head * n_head,
+                                0);
+                    cb(Qcur, "Qcur", il);
+
+                    Qcur = llm_build_norm(ctx0, Qcur, hparams,
+                                model.layers[il].attn_q_norm,
+                                model.layers[il].attn_q_norm_b,
+                                LLM_NORM, cb, il);
+                    cb(Qcur, "Qcur", il);
+                }
+
+                if (model.layers[il].attn_k_norm) {
+                    Kcur = ggml_view_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens,
+                                ggml_element_size(Kcur) * n_embd_head,
+                                ggml_element_size(Kcur) * n_embd_head * n_head_kv,
+                                0);
+                    cb(Kcur, "Kcur", il);
+
+                    Kcur = llm_build_norm(ctx0, Kcur, hparams,
+                               model.layers[il].attn_k_norm,
+                               model.layers[il].attn_k_norm_b,
+                               LLM_NORM, cb, il);
+                    cb(Kcur, "Kcur", il);
                 }
 
                 Qcur = ggml_rope_ext(
@@ -14623,6 +10145,12 @@ struct llm_build_context {
                 cur = llm_build_kv(ctx0, lctx, kv_self, gf,
                         model.layers[il].wo, nullptr,
                         Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
+
+                if (hparams.swin_norm) {
+                    cur = llm_build_norm(ctx0, cur, hparams,
+                        model.layers[il].attn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                }
             }
 
             if (il == n_layer - 1) {
@@ -14637,10 +10165,12 @@ struct llm_build_context {
             cb(ffn_inp, "ffn_inp", il);
 
             // feed-forward network
-            cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    NULL, NULL,
-                    LLM_NORM, cb, il);
-            cb(cur, "ffn_norm", il);
+            if (!hparams.swin_norm) {
+                cur = llm_build_norm(ctx0, ffn_inp, hparams,
+                        model.layers[il].ffn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "ffn_norm", il);
+            }
 
             cur = llm_build_ffn(ctx0, lctx, cur,
                     model.layers[il].ffn_up,   NULL, NULL,
@@ -14650,6 +10180,13 @@ struct llm_build_context {
                     LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
             cb(cur, "ffn_out", il);
 
+            if (hparams.swin_norm) {
+                cur = llm_build_norm(ctx0, cur, hparams,
+                        model.layers[il].ffn_norm, NULL,
+                        LLM_NORM_RMS, cb, il);
+                cb(cur, "ffn_norm", il);
+            }
+
             cur = ggml_add(ctx0, cur, ffn_inp);
             cb(cur, "ffn_out", il);
 
@@ -14663,12 +10200,25 @@ struct llm_build_context {
         cur = inpL;
 
         cur = llm_build_norm(ctx0, cur, hparams,
-                NULL, NULL,
-                LLM_NORM, cb, -1);
+                model.output_norm, NULL,
+                LLM_NORM_RMS, cb, -1);
         cb(cur, "result_norm", -1);
 
         // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_output_with_img_logits", -1);
+
+        // TODO: this suppresses the output of image tokens, which is required to enable text-only outputs.
+        // Needs to be removed once image outputs are supported.
+        int img_token_end_idx = 8196;
+        int img_token_start_idx = 4;
+        int num_img_tokens = img_token_end_idx - img_token_start_idx;
+        // creates 1d tensor of size num_img_tokens and values -FLT_MAX,
+        // which ensures that text token values are always at least larger than image token values
+        struct ggml_tensor * img_logits = ggml_new_tensor_1d(ctx0, GGML_TYPE_F32, num_img_tokens);
+        img_logits = ggml_clamp(ctx0, img_logits, -FLT_MAX, -FLT_MAX);
+        cb(img_logits, "img_logits", -1);
+        cur = ggml_set_1d(ctx0, cur, img_logits, ggml_element_size(cur) * img_token_start_idx);
         cb(cur, "result_output", -1);
 
         ggml_build_forward_expand(gf, cur);
@@ -14676,7216 +10226,1984 @@ struct llm_build_context {
         return gf;
     }
 
-    struct ggml_cgraph * build_olmo2() {
+    struct ggml_cgraph * build_wavtokenizer_dec() {
         struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
 
-        // mutable variable, needed during the last layer of the computation to skip unused tokens
-        int32_t n_tokens = this->n_tokens;
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
-
         struct ggml_tensor * cur;
         struct ggml_tensor * inpL;
 
         inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
 
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
+        cur = ggml_cont(ctx0, ggml_transpose(ctx0, inpL));
 
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
+        cur = ggml_conv_1d_ph(ctx0, model.conv1d, cur, 1, 1);
+        cur = ggml_add(ctx0, cur, model.conv1d_b);
 
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
+        // posnet
+        for (uint32_t il = 0; il < hparams.posnet.n_layer; ++il) {
+            const auto & layer = model.layers[il].posnet;
 
-            cur = inpL;
+            inpL = cur;
 
-            // self_attention
-            {
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
+            switch (il) {
+                case 0:
+                case 1:
+                case 3:
+                case 4:
+                    {
+                        cur = llm_build_norm(ctx0, cur, hparams,
+                                layer.norm1,
+                                layer.norm1_b,
+                                LLM_NORM_GROUP, cb, 0);
 
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
+                        cur = ggml_mul(ctx0, ggml_sigmoid(ctx0, cur), cur);
 
-                Qcur = llm_build_norm(ctx0, Qcur, hparams, model.layers[il].attn_q_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(Qcur, "Qcur_normed", il);
+                        cur = ggml_conv_1d_ph(ctx0, layer.conv1, cur, 1, 1);
+                        cur = ggml_add(ctx0, cur, layer.conv1_b);
 
-                Kcur = llm_build_norm(ctx0, Kcur, hparams, model.layers[il].attn_k_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(Kcur, "Kcur_normed", il);
+                        cur = llm_build_norm(ctx0, cur, hparams,
+                                layer.norm2,
+                                layer.norm2_b,
+                                LLM_NORM_GROUP, cb, 0);
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+                        cur = ggml_mul(ctx0, ggml_sigmoid(ctx0, cur), cur);
 
-                Qcur = ggml_rope_ext(
-                    ctx0, Qcur, inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur_rope", il);
+                        cur = ggml_conv_1d_ph(ctx0, layer.conv2, cur, 1, 1);
+                        cur = ggml_add(ctx0, cur, layer.conv2_b);
 
-                Kcur = ggml_rope_ext(
-                    ctx0, Kcur, inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur_rope", il);
+                        cur = ggml_add(ctx0, cur, inpL);
+                    } break;
+                case 2:
+                    {
+                        cur = llm_build_norm(ctx0, cur, hparams,
+                                layer.attn_norm,
+                                layer.attn_norm_b,
+                                LLM_NORM_GROUP, cb, 0);
 
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-            }
+                        struct ggml_tensor * q;
+                        struct ggml_tensor * k;
+                        struct ggml_tensor * v;
 
-            cur = llm_build_norm(ctx0, cur, hparams,
-                    model.layers[il].attn_post_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_post_norm", il);
+                        q = ggml_conv_1d_ph(ctx0, layer.attn_q, cur, 1, 1);
+                        k = ggml_conv_1d_ph(ctx0, layer.attn_k, cur, 1, 1);
+                        v = ggml_conv_1d_ph(ctx0, layer.attn_v, cur, 1, 1);
 
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                n_tokens = n_outputs;
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
+                        q = ggml_add(ctx0, q, layer.attn_q_b);
+                        k = ggml_add(ctx0, k, layer.attn_k_b);
+                        v = ggml_add(ctx0, v, layer.attn_v_b);
 
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+                        q = ggml_cont(ctx0, ggml_transpose(ctx0, q));
+                        k = ggml_cont(ctx0, ggml_transpose(ctx0, k));
 
-            // feed-forward network
-            cur = llm_build_ffn(ctx0, lctx, ffn_inp,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-            cb(cur, "ffn_out", il);
+                        struct ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
 
-            cur = llm_build_norm(ctx0, cur, hparams,
-                model.layers[il].ffn_post_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-            cb(cur, "ffn_post_norm", -1);
+                        kq = ggml_soft_max_ext(ctx0, kq, nullptr, 1.0f/sqrtf(float(hparams.posnet.n_embd)), 0.0f);
 
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
+                        cur = ggml_mul_mat(ctx0, kq, v);
 
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
+                        cur = ggml_conv_1d_ph(ctx0, layer.attn_o, cur, 1, 1);
+                        cur = ggml_add(ctx0, cur, layer.attn_o_b);
 
-            // input for next layer
-            inpL = cur;
+                        cur = ggml_add(ctx0, cur, inpL);
+                    } break;
+                case 5:
+                    {
+                        cur = llm_build_norm(ctx0, cur, hparams,
+                                layer.norm,
+                                layer.norm_b,
+                                LLM_NORM_GROUP, cb, 0);
+                    } break;
+                default: GGML_ABORT("unknown posnet layer");
+            };
         }
 
-        cur = inpL;
+        cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
 
         cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        // lm_head
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    // based on the build_qwen2moe() function, changes:
-    //   * removed shared experts
-    //   * removed bias
-    //   * added q, k norm
-    struct ggml_cgraph * build_olmoe() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        // mutable variable, needed during the last layer of the computation to skip unused tokens
-        int32_t n_tokens = this->n_tokens;
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
-            // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self_attention
-            {
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = llm_build_norm(ctx0, Qcur, hparams, model.layers[il].attn_q_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(Qcur, "Qcur_normed", il);
+                model.tok_norm,
+                model.tok_norm_b,
+                LLM_NORM, cb, -1);
 
-                Kcur = llm_build_norm(ctx0, Kcur, hparams, model.layers[il].attn_k_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(Kcur, "Kcur_normed", il);
+        cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
 
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
+        inpL = cur;
 
-                Qcur = ggml_rope_ext(
-                    ctx0, Qcur, inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur_rope", il);
+        // convnext
+        for (uint32_t il = 0; il < hparams.convnext.n_layer; ++il) {
+            const auto & layer = model.layers[il].convnext;
 
-                Kcur = ggml_rope_ext(
-                    ctx0, Kcur, inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur_rope", il);
+            cur = inpL;
 
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-            }
+            cur = ggml_conv_1d_dw_ph(ctx0, layer.dw, cur, 1, 1);
+            cur = ggml_add(ctx0, cur, layer.dw_b);
 
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                n_tokens = n_outputs;
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
+            cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
 
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
+            cur = llm_build_norm(ctx0, cur, hparams,
+                    layer.norm,
+                    layer.norm_b,
+                    LLM_NORM, cb, -1);
 
-            // MoE branch
-            cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm", il);
+            cur = llm_build_ffn(ctx0, lctx, cur,
+                    layer.pw1, layer.pw1_b, NULL,
+                    NULL,      NULL,        NULL,
+                    layer.pw2, layer.pw2_b, NULL,
+                    NULL,
+                    LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
 
-            cur = llm_build_moe_ffn(ctx0, lctx, cur,
-                    model.layers[il].ffn_gate_inp,
-                    model.layers[il].ffn_up_exps,
-                    model.layers[il].ffn_gate_exps,
-                    model.layers[il].ffn_down_exps,
-                    n_expert, n_expert_used,
-                    LLM_FFN_SILU, false,
-                    false, 0.0,
-                    cb, il);
-            cb(cur, "ffn_moe_out", il);
+            cur = ggml_mul(ctx0, cur, layer.gamma);
 
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
+            cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
 
-            // input for next layer
-            inpL = cur;
+            inpL = ggml_add(ctx0, cur, inpL);
         }
 
         cur = inpL;
 
+        cur = ggml_cont(ctx0, ggml_transpose(ctx0, cur));
+
         cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
+                model.output_norm,
+                model.output_norm_b,
+                LLM_NORM, cb, -1);
 
         // lm_head
         cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
+
+        cur = ggml_add(ctx0, cur, model.output_b);
+        cb(cur, "result_embd", -1);
 
         ggml_build_forward_expand(gf, cur);
 
         return gf;
     }
+};
 
-    struct ggml_cgraph * build_openelm() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
+static struct ggml_cgraph * llama_build_graph_defrag(llama_context & lctx, const std::vector & ids) {
+    llama_ubatch dummy = {};
+    dummy.equal_seqs = true;
 
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
+    llm_build_cb cb = [&](struct ggml_tensor * , const char * , int ) { };
 
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
+    struct llm_build_context llm(lctx, dummy, cb, false);
 
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
+    llm.init();
 
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
+    struct ggml_cgraph * result = llm.build_defrag(ids);
 
-        for (int il = 0; il < n_layer; ++il) {
-            const int64_t n_head    = hparams.n_head(il);
-            const int64_t n_head_kv = hparams.n_head_kv(il);
-            const int64_t n_head_qkv = 2*n_head_kv + n_head;
+    llm.free();
 
-            cur = inpL;
-            struct ggml_tensor * residual = cur;
+    return result;
+}
 
-            // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
+static struct ggml_cgraph * llama_build_graph_k_shift(llama_context & lctx) {
+    llama_ubatch dummy = {};
+    dummy.equal_seqs = true;
 
-            // self-attention
-            {
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
+    llm_build_cb cb = [&](struct ggml_tensor * , const char * , int ) { };
 
-                cur = ggml_reshape_3d(ctx0, cur, n_embd_head_k, n_head_qkv, n_tokens);
+    struct llm_build_context llm(lctx, dummy, cb, false);
 
-                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd_head, n_head, n_tokens, cur->nb[1], cur->nb[2], 0));
-                cb(Qcur, "Qcur", il);
+    llm.init();
 
-                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, cur->nb[1], cur->nb[2], cur->nb[1]*n_head));
-                cb(Kcur, "Kcur", il);
+    struct ggml_cgraph * result = llm.build_k_shift();
 
-                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd_head, n_head_kv, n_tokens, cur->nb[1], cur->nb[2], cur->nb[1]*(n_head+n_head_kv)));
-                cb(Vcur, "Vcur", il);
+    llm.free();
 
-                Qcur = llm_build_norm(ctx0, Qcur, hparams,
-                        model.layers[il].attn_q_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(Qcur, "Qcur", il);
-
-                Kcur = llm_build_norm(ctx0, Kcur, hparams,
-                        model.layers[il].attn_k_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(Kcur, "Kcur", il);
-
-                Qcur = ggml_rope_ext(
-                    ctx0, Qcur, inp_pos, NULL, n_rot, rope_type, n_ctx_orig,
-                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur", il);
-
-                Kcur = ggml_rope_ext(
-                    ctx0, Kcur, inp_pos, NULL, n_rot, rope_type, n_ctx_orig,
-                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur", il);
-
-                Vcur = ggml_reshape_2d(ctx0, Vcur, n_embd_head * n_head_kv, n_tokens);
-                cb(Qcur, "Vcur", il);
-
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                residual = ggml_get_rows(ctx0, residual, inp_out_ids);
-                cur = ggml_get_rows(ctx0, cur, inp_out_ids);
-            }
-
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, residual, cur);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-                cb(cur, "ffn_out", il);
-            }
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
-
-            inpL = cur;
-        }
-
-        cur = inpL;
-
-        // norm
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    struct ggml_cgraph * build_gptneox() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                cb(cur, "bqkv", il);
-
-                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
-                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur", il);
-
-                Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur", il);
-
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
-            }
-
-            // ffn
-            if (hparams.use_par_res) {
-                // attention and ffn are computed in parallel
-                // x = x + attn(ln1(x)) + ffn(ln2(x))
-
-                struct ggml_tensor * attn_out = cur;
-
-                cur = llm_build_norm(ctx0, inpL, hparams,
-                        model.layers[il].ffn_norm,
-                        model.layers[il].ffn_norm_b,
-                        LLM_NORM, cb, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        NULL,                      NULL,                        NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                        NULL,
-                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
-                cb(cur, "ffn_out", il);
-
-                cur = ggml_add(ctx0, cur, inpL);
-                cb(cur, "ffn_out", il);
-
-                cur = ggml_add(ctx0, cur, attn_out);
-                cur = lctx.cvec.apply_to(ctx0, cur, il);
-                cb(cur, "l_out", il);
-
-                // input for next layer
-                inpL = cur;
-            } else {
-                // attention and ffn are computed sequentially
-                // x = x + attn(ln1(x))
-                // x = x + ffn(ln2(x))
-
-                struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
-                cb(ffn_inp, "ffn_inp", il);
-
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm,
-                        model.layers[il].ffn_norm_b,
-                        LLM_NORM, cb, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        NULL,                      NULL,                        NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                        NULL,
-                        LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
-                cb(cur, "ffn_out", il);
-
-                cur = ggml_add(ctx0, cur, ffn_inp);
-                cur = lctx.cvec.apply_to(ctx0, cur, il);
-                cb(cur, "l_out", il);
-
-                // input for next layer
-                inpL = cur;
-            }
-        }
-
-        cur = llm_build_norm(ctx0, inpL, hparams,
-                model.output_norm,
-                model.output_norm_b,
-                LLM_NORM, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    struct ggml_cgraph * build_arctic() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        // mutable variable, needed during the last layer of the computation to skip unused tokens
-        int32_t n_tokens = this->n_tokens;
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
-            // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur", il);
-
-                Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur", il);
-
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                n_tokens = n_outputs;
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = llm_build_ffn(ctx0, lctx, cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-            cb(cur, "ffn_out", il);
-
-            struct ggml_tensor * ffn_out = ggml_add(ctx0, cur, ffn_inp);
-            cb(ffn_out, "ffn_out", il);
-
-            // MoE
-            cur = llm_build_norm(ctx0, inpSA, hparams,
-                    model.layers[il].ffn_norm_exps, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm_exps", il);
-
-            cur = llm_build_moe_ffn(ctx0, lctx, cur,
-                    model.layers[il].ffn_gate_inp,
-                    model.layers[il].ffn_up_exps,
-                    model.layers[il].ffn_gate_exps,
-                    model.layers[il].ffn_down_exps,
-                    n_expert, n_expert_used,
-                    LLM_FFN_SILU, true,
-                    false, 0.0,
-                    cb, il);
-            cb(cur, "ffn_moe_out", il);
-
-            cur = ggml_add(ctx0, cur, ffn_out);
-            cb(cur, "ffn_out", il);
-
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        // lm_head
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    struct ggml_cgraph * build_deepseek() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        // mutable variable, needed during the last layer of the computation to skip unused tokens
-        int32_t n_tokens = this->n_tokens;
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-        const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
-            // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                // rope freq factors for llama3; may return nullptr for llama2 and other models
-                struct ggml_tensor * rope_factors = build_rope_factors(il);
-
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-
-                Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, rope_factors,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur", il);
-
-                Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, rope_factors,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur", il);
-
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, kq_scale, cb, il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                n_tokens = n_outputs;
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-
-
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm", il);
-
-            if ((uint32_t) il < hparams.n_layer_dense_lead) {
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-                cb(cur, "ffn_out", il);
-            } else {
-                // MoE branch
-                ggml_tensor * moe_out =
-                        llm_build_moe_ffn(ctx0, lctx, cur,
-                            model.layers[il].ffn_gate_inp,
-                            model.layers[il].ffn_up_exps,
-                            model.layers[il].ffn_gate_exps,
-                            model.layers[il].ffn_down_exps,
-                            n_expert, n_expert_used,
-                            LLM_FFN_SILU, false,
-                            false, hparams.expert_weights_scale,
-                            cb, il);
-                cb(moe_out, "ffn_moe_out", il);
-
-                // FFN shared expert
-                {
-                    ggml_tensor * ffn_shexp = llm_build_ffn(ctx0, lctx, cur,
-                            model.layers[il].ffn_up_shexp,   NULL, NULL,
-                            model.layers[il].ffn_gate_shexp, NULL, NULL,
-                            model.layers[il].ffn_down_shexp, NULL, NULL,
-                            NULL,
-                            LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-                    cb(ffn_shexp, "ffn_shexp", il);
-
-                    cur = ggml_add(ctx0, moe_out, ffn_shexp);
-                    cb(cur, "ffn_out", il);
-                }
-            }
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        // lm_head
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    struct ggml_cgraph * build_deepseek2() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        // mutable variable, needed during the last layer of the computation to skip unused tokens
-        int32_t n_tokens = this->n_tokens;
-
-        bool is_lite = (hparams.n_layer == 27);
-
-        // We have to pre-scale kq_scale and attn_factor to make the YaRN RoPE work correctly.
-        // See https://github.com/ggerganov/llama.cpp/discussions/7416 for detailed explanation.
-        const float mscale = attn_factor * (1.0f + hparams.rope_yarn_log_mul * logf(1.0f / freq_scale));
-        const float kq_scale = 1.0f*mscale*mscale/sqrtf(float(hparams.n_embd_head_k));
-        const float attn_factor_scaled = 1.0f / (1.0f + 0.1f * logf(1.0f / freq_scale));
-
-        const uint32_t n_embd_head_qk_rope = hparams.n_rot;
-        const uint32_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot;
-        const uint32_t kv_lora_rank = hparams.n_lora_kv;
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        // {n_embd, n_tokens}
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
-            // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self_attention
-            {
-                struct ggml_tensor * q = NULL;
-                if (!is_lite) {
-                    // {n_embd, q_lora_rank} * {n_embd, n_tokens} -> {q_lora_rank, n_tokens}
-                    q = ggml_mul_mat(ctx0, model.layers[il].wq_a, cur);
-                    cb(q, "q", il);
-
-                    q = llm_build_norm(ctx0, q, hparams,
-                            model.layers[il].attn_q_a_norm, NULL,
-                            LLM_NORM_RMS, cb, il);
-                    cb(q, "q", il);
-
-                    // {q_lora_rank, n_head * hparams.n_embd_head_k} * {q_lora_rank, n_tokens} -> {n_head * hparams.n_embd_head_k, n_tokens}
-                    q = ggml_mul_mat(ctx0, model.layers[il].wq_b, q);
-                    cb(q, "q", il);
-                } else {
-                    q = ggml_mul_mat(ctx0, model.layers[il].wq, cur);
-                    cb(q, "q", il);
-                }
-
-                // split into {n_head * n_embd_head_qk_nope, n_tokens}
-                struct ggml_tensor * q_nope = ggml_view_3d(ctx0, q, n_embd_head_qk_nope, n_head, n_tokens,
-                        ggml_row_size(q->type, hparams.n_embd_head_k),
-                        ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
-                        0);
-                cb(q_nope, "q_nope", il);
-
-                // and {n_head * n_embd_head_qk_rope, n_tokens}
-                struct ggml_tensor * q_pe = ggml_view_3d(ctx0, q, n_embd_head_qk_rope, n_head, n_tokens,
-                        ggml_row_size(q->type, hparams.n_embd_head_k),
-                        ggml_row_size(q->type, hparams.n_embd_head_k * n_head),
-                        ggml_row_size(q->type, n_embd_head_qk_nope));
-                cb(q_pe, "q_pe", il);
-
-                // {n_embd, kv_lora_rank + n_embd_head_qk_rope} * {n_embd, n_tokens} -> {kv_lora_rank + n_embd_head_qk_rope, n_tokens}
-                struct ggml_tensor * kv_pe_compresseed = ggml_mul_mat(ctx0, model.layers[il].wkv_a_mqa, cur);
-                cb(kv_pe_compresseed, "kv_pe_compresseed", il);
-
-                // split into {kv_lora_rank, n_tokens}
-                struct ggml_tensor * kv_compressed = ggml_view_2d(ctx0, kv_pe_compresseed, kv_lora_rank, n_tokens,
-                        kv_pe_compresseed->nb[1],
-                        0);
-                cb(kv_compressed, "kv_compressed", il);
-
-                // and {n_embd_head_qk_rope, n_tokens}
-                struct ggml_tensor * k_pe = ggml_view_3d(ctx0, kv_pe_compresseed, n_embd_head_qk_rope, 1, n_tokens,
-                        kv_pe_compresseed->nb[1],
-                        kv_pe_compresseed->nb[1],
-                        ggml_row_size(kv_pe_compresseed->type, kv_lora_rank));
-                cb(k_pe, "k_pe", il);
-
-                kv_compressed = ggml_cont(ctx0, kv_compressed); // TODO: the CUDA backend does not support non-contiguous norm
-                kv_compressed = llm_build_norm(ctx0, kv_compressed, hparams,
-                        model.layers[il].attn_kv_a_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(kv_compressed, "kv_compressed", il);
-
-                // {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)} * {kv_lora_rank, n_tokens} -> {n_head * (n_embd_head_qk_nope + n_embd_head_v), n_tokens}
-                struct ggml_tensor * kv = ggml_mul_mat(ctx0, model.layers[il].wkv_b, kv_compressed);
-                cb(kv, "kv", il);
-
-                // split into {n_head * n_embd_head_qk_nope, n_tokens}
-                struct ggml_tensor * k_nope = ggml_view_3d(ctx0, kv, n_embd_head_qk_nope, n_head, n_tokens,
-                        ggml_row_size(kv->type, n_embd_head_qk_nope + hparams.n_embd_head_v),
-                        ggml_row_size(kv->type, n_head * (n_embd_head_qk_nope + hparams.n_embd_head_v)),
-                        0);
-                cb(k_nope, "k_nope", il);
-
-                // and {n_head * n_embd_head_v, n_tokens}
-                struct ggml_tensor * v_states = ggml_view_3d(ctx0, kv, hparams.n_embd_head_v, n_head, n_tokens,
-                        ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)),
-                        ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)*n_head),
-                        ggml_row_size(kv->type, (n_embd_head_qk_nope)));
-                cb(v_states, "v_states", il);
-
-                v_states = ggml_cont(ctx0, v_states);
-                cb(v_states, "v_states", il);
-
-                v_states = ggml_view_2d(ctx0, v_states, hparams.n_embd_head_v * n_head, n_tokens,
-                    ggml_row_size(kv->type, hparams.n_embd_head_v * n_head),
-                    0);
-                cb(v_states, "v_states", il);
-
-                q_pe = ggml_cont(ctx0, q_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
-                q_pe = ggml_rope_ext(
-                    ctx0, q_pe, inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor_scaled, beta_fast, beta_slow
-                );
-                cb(q_pe, "q_pe", il);
-
-                // shared RoPE key
-                k_pe = ggml_cont(ctx0, k_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
-                k_pe = ggml_rope_ext(
-                    ctx0, k_pe, inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor_scaled, beta_fast, beta_slow
-                );
-                cb(k_pe, "k_pe", il);
-
-                struct ggml_tensor * q_states = ggml_concat(ctx0, q_nope, q_pe, 0);
-                cb(q_states, "q_states", il);
-
-                struct ggml_tensor * k_states = ggml_concat(ctx0, k_nope, ggml_repeat(ctx0, k_pe, q_pe), 0);
-                cb(k_states, "k_states", il);
-
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
-                        k_states, v_states, q_states, KQ_mask, n_tokens, kv_head, n_kv, kq_scale, cb, il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                n_tokens = n_outputs;
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm", il);
-
-            if ((uint32_t) il < hparams.n_layer_dense_lead) {
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-                cb(cur, "ffn_out", il);
-            } else {
-                // MoE branch
-                ggml_tensor * moe_out =
-                        llm_build_moe_ffn(ctx0, lctx, cur,
-                            model.layers[il].ffn_gate_inp,
-                            model.layers[il].ffn_up_exps,
-                            model.layers[il].ffn_gate_exps,
-                            model.layers[il].ffn_down_exps,
-                            n_expert, n_expert_used,
-                            LLM_FFN_SILU, false,
-                            true, hparams.expert_weights_scale,
-                            cb, il);
-                cb(moe_out, "ffn_moe_out", il);
-
-                // FFN shared expert
-                {
-                    ggml_tensor * ffn_shexp = llm_build_ffn(ctx0, lctx, cur,
-                            model.layers[il].ffn_up_shexp,   NULL, NULL,
-                            model.layers[il].ffn_gate_shexp, NULL, NULL,
-                            model.layers[il].ffn_down_shexp, NULL, NULL,
-                            NULL,
-                            LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-                    cb(ffn_shexp, "ffn_shexp", il);
-
-                    cur = ggml_add(ctx0, moe_out, ffn_shexp);
-                    cb(cur, "ffn_out", il);
-                }
-            }
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        // lm_head
-        cur = ggml_mul_mat(ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    struct ggml_cgraph * build_bitnet() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                if (model.layers[il].wq_scale) {
-                    Qcur = ggml_mul(ctx0, Qcur, model.layers[il].wq_scale);
-                }
-                cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-
-                // B1.K
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                if (model.layers[il].wk_scale) {
-                    Kcur = ggml_mul(ctx0, Kcur, model.layers[il].wk_scale);
-                }
-                cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-
-                // B1.V
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
-                if (model.layers[il].wv_scale) {
-                    Vcur = ggml_mul(ctx0, Vcur, model.layers[il].wv_scale);
-                }
-                cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-
-                Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur", il);
-
-                Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur", il);
-
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        NULL, NULL,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-
-                cur = llm_build_norm(ctx0, cur, hparams,
-                        model.layers[il].attn_sub_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "attn_sub_norm", il);
-
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo, cur);
-                if (model.layers[il].wo_scale) {
-                    cur = ggml_mul(ctx0, cur, model.layers[il].wo_scale);
-                }
-                if (model.layers[il].bo) {
-                    cur = ggml_add(ctx0, cur, model.layers[il].bo);
-                }
-                cb(cur, "attn_o_out", il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward forward
-            cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = llm_build_ffn(ctx0, lctx, cur,
-                    model.layers[il].ffn_up,   NULL, model.layers[il].ffn_up_scale,
-                    model.layers[il].ffn_gate, NULL, model.layers[il].ffn_gate_scale,
-                    NULL,                      NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-            cb(cur, "ffn_sub_out", il);
-
-            cur = llm_build_norm(ctx0, cur, hparams,
-                            model.layers[il].ffn_sub_norm, NULL,
-                            LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_sub_norm", il);
-
-            cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].ffn_down, cur);
-            if (model.layers[il].ffn_down_scale) {
-                cur = ggml_mul(ctx0, cur, model.layers[il].ffn_down_scale);
-            }
-            cb(cur, "ffn_down", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        // lm_head
-        // FIXME: do not use model.tok_embd directly, duplicate as model.output
-        cur = llm_build_lora_mm(lctx, ctx0, model.tok_embd, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-        return gf;
-    }
-
-    struct ggml_cgraph * build_t5_encoder() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        // mutable variable, needed during the last layer of the computation to skip unused tokens
-        int32_t n_tokens = this->n_tokens;
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        GGML_ASSERT(lctx.is_encoding);
-        struct ggml_tensor * pos_bucket_enc = llm_build_pos_bucket(false);
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask_enc = build_inp_KQ_mask(false);
-
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
-            // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm_enc, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq_enc, cur);
-                cb(Qcur, "Qcur", il);
-
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk_enc, cur);
-                cb(Kcur, "Kcur", il);
-
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv_enc, cur);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
-
-                struct ggml_tensor * q =                 ggml_permute(ctx0, Qcur, 0, 2, 1, 3);
-                struct ggml_tensor * k = ggml_cont(ctx0, ggml_permute(ctx0, Kcur, 0, 2, 1, 3));
-
-                struct ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
-                cb(kq, "kq", il);
-
-                struct ggml_tensor * attn_rel_b = model.layers[il].attn_rel_b_enc ? model.layers[il].attn_rel_b_enc : model.layers[0].attn_rel_b_enc;
-                struct ggml_tensor * pos_bias = llm_build_pos_bias(pos_bucket_enc, attn_rel_b);
-                struct ggml_tensor * kq_b = ggml_add(ctx0, kq, pos_bias);
-                cb(kq_b, "kq_b", il);
-
-                kq = ggml_soft_max_ext(ctx0, kq_b, KQ_mask_enc, 1.0f, hparams.f_max_alibi_bias);
-                cb(kq, "kq_soft_max_ext", il);
-
-                struct ggml_tensor * v = ggml_cont(ctx0, ggml_transpose(ctx0, ggml_reshape_2d(ctx0, Vcur, n_embd_gqa, n_tokens)));
-                cb(v, "v", il);
-
-                struct ggml_tensor * kqv = ggml_mul_mat(ctx0, ggml_reshape_3d(ctx0, v, n_tokens, n_embd_head, n_head_kv), kq);
-                cb(kqv, "kqv", il);
-
-                struct ggml_tensor * kqv_merged = ggml_permute(ctx0, kqv, 0, 2, 1, 3);
-                cb(kqv_merged, "kqv_merged", il);
-
-                cur = ggml_cont_2d(ctx0, kqv_merged, n_embd_gqa, n_tokens);
-                cb(cur, "kqv_merged_cont", il);
-
-                ggml_build_forward_expand(gf, cur);
-
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo_enc, cur);
-                cb(cur, "kqv_out", il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                n_tokens = n_outputs;
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm_enc, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "ffn_norm", il);
-
-                // T5 uses relu, flan-T5 uses gelu-gated
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up_enc,   NULL, NULL,
-                        model.layers[il].ffn_gate_enc, NULL, NULL,
-                        model.layers[il].ffn_down_enc, NULL, NULL,
-                        NULL,
-                        model.layers[il].ffn_gate_enc ? LLM_FFN_GELU : LLM_FFN_RELU,
-                        model.layers[il].ffn_gate_enc ? LLM_FFN_PAR  : LLM_FFN_SEQ,
-                        cb, il);
-                cb(cur, "ffn_out", il);
-            }
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            ggml_tensor * layer_dir = lctx.cvec.tensor_for(il);
-            if (layer_dir != nullptr) {
-                cur = ggml_add(ctx0, cur, layer_dir);
-            }
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        cur = inpL;
-        cb(cur, "result_embd", -1);
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm_enc, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    struct ggml_cgraph * build_t5_decoder() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        // mutable variable, needed during the last layer of the computation to skip unused tokens
-        int32_t n_tokens = this->n_tokens;
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        GGML_ASSERT(!lctx.is_encoding);
-        GGML_ASSERT(n_outputs_enc > 0 && "call llama_encode() first");
-
-        struct ggml_tensor * embd_enc       = llm_build_inp_embd_enc();
-        struct ggml_tensor * pos_bucket_dec = llm_build_pos_bucket(true);
-
-        struct ggml_tensor * KQ_mask_dec   = build_inp_KQ_mask();
-        struct ggml_tensor * KQ_mask_cross = llm_build_inp_KQ_mask_cross();
-
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
-            // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-
-                llm_build_kv_store(ctx0, hparams, cparams, kv_self, gf, Kcur, Vcur, n_tokens, kv_head, cb, il);
-
-                struct ggml_tensor * k =
-                    ggml_view_3d(ctx0, kv_self.k_l[il],
-                            n_embd_head_k, n_kv, n_head_kv,
-                            ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa),
-                            ggml_row_size(kv_self.k_l[il]->type, n_embd_head_k),
-                            0);
-                cb(k, "k", il);
-
-                struct ggml_tensor * v =
-                    ggml_view_3d(ctx0, kv_self.v_l[il],
-                            n_kv, n_embd_head_v, n_head_kv,
-                            ggml_element_size(kv_self.v_l[il])*n_ctx,
-                            ggml_element_size(kv_self.v_l[il])*n_ctx*n_embd_head_v,
-                            0);
-                cb(v, "v", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
-
-                struct ggml_tensor * q = ggml_permute(ctx0, Qcur, 0, 2, 1, 3);
-
-                struct ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
-                cb(kq, "kq", il);
-
-                struct ggml_tensor * attn_rel_b = model.layers[il].attn_rel_b ? model.layers[il].attn_rel_b : model.layers[0].attn_rel_b;
-                struct ggml_tensor * pos_bias = llm_build_pos_bias(pos_bucket_dec, attn_rel_b);
-                struct ggml_tensor * kq_b = ggml_add(ctx0, kq, pos_bias);
-                cb(kq_b, "kq_b", il);
-
-                kq = ggml_soft_max_ext(ctx0, kq_b, KQ_mask_dec, 1.0f, hparams.f_max_alibi_bias);
-                cb(kq, "kq_soft_max_ext", il);
-
-                struct ggml_tensor * kqv = ggml_mul_mat(ctx0, v, kq);
-                cb(kqv, "kqv", il);
-
-                struct ggml_tensor * kqv_merged = ggml_permute(ctx0, kqv, 0, 2, 1, 3);
-                cb(kqv_merged, "kqv_merged", il);
-
-                cur = ggml_cont_2d(ctx0, kqv_merged, n_embd_gqa, n_tokens);
-                cb(cur, "kqv_merged_cont", il);
-
-                ggml_build_forward_expand(gf, cur);
-
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo, cur);
-                cb(cur, "kqv_out", il);
-            }
-
-            cur = ggml_add(ctx0, cur, inpSA);
-            cb(cur, "cross_inp", il);
-
-            struct ggml_tensor * inpCA = cur;
-
-            // norm
-            cur = llm_build_norm(ctx0, cur, hparams,
-                    model.layers[il].attn_norm_cross, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm_cross", il);
-
-            // cross-attention
-            {
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq_cross, cur);
-                cb(Qcur, "Qcur", il);
-
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk_cross, embd_enc);
-                cb(Kcur, "Kcur", il);
-
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv_cross, embd_enc);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);
-                Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_outputs_enc);
-
-                struct ggml_tensor * q =                 ggml_permute(ctx0, Qcur, 0, 2, 1, 3);
-                struct ggml_tensor * k = ggml_cont(ctx0, ggml_permute(ctx0, Kcur, 0, 2, 1, 3));
-
-                struct ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
-                cb(kq, "kq", il);
-
-                kq = ggml_soft_max_ext(ctx0, kq, KQ_mask_cross, 1.0f, hparams.f_max_alibi_bias);
-                cb(kq, "kq_soft_max_ext", il);
-
-                struct ggml_tensor * v = ggml_cont(ctx0, ggml_transpose(ctx0, ggml_reshape_2d(ctx0, Vcur, n_embd_gqa, n_outputs_enc)));
-                cb(v, "v", il);
-
-                struct ggml_tensor * kqv = ggml_mul_mat(ctx0, ggml_reshape_3d(ctx0, v, n_outputs_enc, n_embd_head, n_head_kv), kq);
-                cb(kqv, "kqv", il);
-
-                struct ggml_tensor * kqv_merged = ggml_permute(ctx0, kqv, 0, 2, 1, 3);
-                cb(kqv_merged, "kqv_merged", il);
-
-                cur = ggml_cont_2d(ctx0, kqv_merged, n_embd_gqa, n_tokens);
-                cb(cur, "kqv_merged_cont", il);
-
-                ggml_build_forward_expand(gf, cur);
-
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo_cross, cur);
-                cb(cur, "kqv_out", il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                n_tokens = n_outputs;
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-                inpCA = ggml_get_rows(ctx0, inpCA, inp_out_ids);
-            }
-
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpCA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "ffn_norm", il);
-
-                // T5 uses relu, flan-T5 uses gelu-gated
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        model.layers[il].ffn_gate, NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        model.layers[il].ffn_gate_enc ? LLM_FFN_GELU : LLM_FFN_RELU,
-                        model.layers[il].ffn_gate_enc ? LLM_FFN_PAR : LLM_FFN_SEQ,
-                        cb, il);
-                cb(cur, "ffn_out", il);
-            }
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            ggml_tensor * layer_dir = lctx.cvec.tensor_for(il);
-            if (layer_dir != nullptr) {
-                cur = ggml_add(ctx0, cur, layer_dir);
-            }
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        cur = inpL;
-        cb(cur, "result_embd", -1);
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        // lm_head
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    struct ggml_cgraph * build_jais() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                cb(cur, "bqkv", il);
-
-                struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*cur->nb[0]*(n_embd)));
-                struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*cur->nb[0]*(n_embd)));
-                struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*cur->nb[0]*(n_embd + n_embd_gqa)));
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-
-                Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
-
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/float(n_embd_head), cb, il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);
-                inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);
-            }
-
-            // add the input
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // FF
-            {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm,
-                        model.layers[il].ffn_norm_b,
-                        LLM_NORM, cb, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                        model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                        NULL,
-                        LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-                cb(cur, "ffn_out", il);
-            }
-
-            inpL = ggml_add(ctx0, cur, ffn_inp);
-            cb(inpL, "l_out", il);
-        }
-
-        cur = llm_build_norm(ctx0, inpL, hparams,
-                model.output_norm,
-                model.output_norm_b,
-                LLM_NORM, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    struct ggml_cgraph * build_chatglm() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        const int64_t n_embd_gqa  = hparams.n_embd_v_gqa();
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                struct ggml_tensor * Qcur = nullptr;
-                struct ggml_tensor * Kcur = nullptr;
-                struct ggml_tensor * Vcur = nullptr;
-
-                cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                cb(cur, "bqkv", il);
-
-                Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd,     n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
-                Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
-                Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
-
-                cb(Qcur, "Qcur", il);
-                cb(Kcur, "Kcur", il);
-                cb(Vcur, "Vcur", il);
-                //printf("freq_base: %f freq_scale: %f ext_factor: %f attn_factor: %f\n", freq_base, freq_scale, ext_factor, attn_factor);
-                Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur_rope", il);
-
-                Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur_rope", il);
-
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, NULL,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-
-            // Add the input
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // FF
-            {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm,
-                        NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = llm_build_ffn(ctx0, lctx, cur,
-                        model.layers[il].ffn_up,   NULL, NULL,
-                        NULL,                      NULL, NULL,
-                        model.layers[il].ffn_down, NULL, NULL,
-                        NULL,
-                        LLM_FFN_SWIGLU, LLM_FFN_SEQ, cb, il);
-                cb(cur, "ffn_out", il);
-
-            }
-
-            inpL = ggml_add(ctx0, cur, ffn_inp);
-            cb(inpL, "l_out", il);
-        }
-
-        cur = llm_build_norm(ctx0, inpL, hparams,
-                model.output_norm,
-                NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    struct ggml_cgraph * build_nemotron() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        //GGML_ASSERT(n_embd_head == hparams.n_rot);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
-            // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-
-                Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur", il);
-
-                Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur", il);
-
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    model.layers[il].ffn_norm,
-                    model.layers[il].ffn_norm_b,
-                    LLM_NORM, cb, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = llm_build_ffn(ctx0, lctx, cur,
-                    model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,
-                    NULL,                      NULL,                        NULL,
-                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
-                    NULL,
-                    LLM_FFN_RELU_SQR, LLM_FFN_SEQ, cb, il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, model.output_norm_b,
-                LLM_NORM, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        // lm_head
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    struct ggml_cgraph * build_exaone() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        // mutable variable, needed during the last layer of the computation to skip unused tokens
-        int32_t n_tokens = this->n_tokens;
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
-            // norm
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self-attention
-            {
-                // rope freq factors for llama3; may return nullptr for llama2 and other models
-                struct ggml_tensor * rope_factors = build_rope_factors(il);
-
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-                if (model.layers[il].bq) {
-                    Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
-                    cb(Qcur, "Qcur", il);
-                }
-
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-                if (model.layers[il].bk) {
-                    Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);
-                    cb(Kcur, "Kcur", il);
-                }
-
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-                if (model.layers[il].bv) {
-                    Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);
-                    cb(Vcur, "Vcur", il);
-                }
-
-                Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, rope_factors,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur", il);
-
-                Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, rope_factors,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur", il);
-
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                n_tokens = n_outputs;
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                    model.layers[il].ffn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-            cb(cur, "ffn_norm", il);
-
-            cur = llm_build_ffn(ctx0, lctx, cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-            cb(cur, "ffn_out", il);
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        // lm_head
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    ggml_cgraph * build_rwkv6() {
-        ggml_cgraph *gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        // Token shift state dimensions should be 2 * n_emb
-        GGML_ASSERT(n_embd == hparams.n_embd_k_s() / 2);
-
-        const int64_t n_seqs = ubatch.n_seqs;
-        const int64_t n_seq_tokens = ubatch.n_seq_tokens;
-        const int64_t n_tokens = ubatch.n_tokens;
-        GGML_ASSERT(n_seqs != 0);
-        GGML_ASSERT(ubatch.equal_seqs);
-        GGML_ASSERT(n_tokens == n_seq_tokens * n_seqs);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-        struct ggml_tensor * state_copy = build_inp_s_copy();
-        struct ggml_tensor * state_mask = build_inp_s_mask();
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-        inpL = llm_build_norm(ctx0, inpL, hparams, model.tok_norm, model.tok_norm_b, LLM_NORM, cb, -1);
-
-        for (int il = 0; il < n_layer; ++il) {
-            const llama_layer * layer = &model.layers[il];
-
-            // (ab)using the KV cache to store the states
-            struct ggml_tensor * token_shift = llm_build_copy_mask_state(ctx0,
-                    gf, kv_self.k_l[il], state_copy, state_mask,
-                    hparams.n_embd_k_s(), kv_self.size, kv_head, n_kv, n_seqs);
-            struct ggml_tensor * wkv_states = llm_build_copy_mask_state(ctx0,
-                    gf, kv_self.v_l[il], state_copy, state_mask,
-                    hparams.n_embd_v_s(), kv_self.size, kv_head, n_kv, n_seqs);
-
-            cur = ggml_reshape_3d(ctx0, inpL, n_embd, n_seq_tokens, n_seqs);
-            token_shift = ggml_reshape_3d(ctx0, token_shift, n_embd, 2, n_seqs);
-
-            struct ggml_tensor * att_shift = ggml_view_3d(ctx0, token_shift, n_embd, 1, n_seqs, token_shift->nb[1], token_shift->nb[2], 0);
-            struct ggml_tensor * ffn_shift = ggml_view_3d(ctx0, token_shift, n_embd, 1, n_seqs, token_shift->nb[1], token_shift->nb[2], n_embd * ggml_element_size(token_shift));
-
-            struct ggml_tensor * x_norm_att = llm_build_norm(ctx0, cur, hparams, layer->attn_norm, layer->attn_norm_b, LLM_NORM, cb, il);
-            struct ggml_tensor * x_prev = ggml_concat(
-                ctx0,
-                att_shift,
-                ggml_view_3d(ctx0, x_norm_att, n_embd, n_seq_tokens - 1, n_seqs, x_norm_att->nb[1], x_norm_att->nb[2], 0),
-                1
-            );
-
-            cur = ggml_add(ctx0, cur, llm_build_rwkv6_time_mix(lctx, ctx0, layer, x_norm_att, x_prev, &wkv_states));
-            ggml_build_forward_expand(gf, cur);
-            ggml_build_forward_expand(
-                gf,
-                ggml_cpy(
-                    ctx0,
-                    wkv_states,
-                    ggml_view_1d(
-                        ctx0,
-                        kv_self.v_l[il],
-                        hparams.n_embd_v_s() * n_seqs,
-                        hparams.n_embd_v_s() * kv_head * ggml_element_size(kv_self.v_l[il])
-                    )
-                )
-            );
-
-            struct ggml_tensor * x_norm_ffn = llm_build_norm(ctx0, cur, hparams, layer->attn_norm_2, layer->attn_norm_2_b, LLM_NORM, cb, il);
-            x_prev = ggml_concat(
-                ctx0,
-                ffn_shift,
-                ggml_view_3d(ctx0, x_norm_ffn, n_embd, n_seq_tokens - 1, n_seqs, x_norm_ffn->nb[1], x_norm_ffn->nb[2], 0),
-                1
-            );
-            cur = ggml_add(ctx0, cur, llm_build_rwkv6_channel_mix(lctx, ctx0, layer, x_norm_ffn, x_prev));
-            ggml_build_forward_expand(gf, cur);
-
-            struct ggml_tensor * last_norm_att = ggml_view_3d(ctx0, x_norm_att, n_embd, 1, n_seqs, x_norm_att->nb[1], x_norm_att->nb[2], (n_seq_tokens-1)*n_embd*ggml_element_size(x_norm_att));
-            struct ggml_tensor * last_norm_ffn = ggml_view_3d(ctx0, x_norm_ffn, n_embd, 1, n_seqs, x_norm_ffn->nb[1], x_norm_ffn->nb[2], (n_seq_tokens-1)*n_embd*ggml_element_size(x_norm_ffn));
-
-            token_shift = ggml_concat(ctx0, last_norm_att, last_norm_ffn, 1);
-
-            ggml_build_forward_expand(
-                gf,
-                ggml_cpy(
-                    ctx0,
-                    ggml_view_1d(ctx0, token_shift, n_embd * n_seqs * 2, 0),
-                    ggml_view_1d(ctx0, kv_self.k_l[il], hparams.n_embd_k_s() * n_seqs, hparams.n_embd_k_s() * kv_head * ggml_element_size(kv_self.k_l[il]))
-                )
-            );
-
-            if (hparams.rescale_every_n_layers != 0 && (il + 1) % hparams.rescale_every_n_layers == 0) {
-                cur = ggml_scale(ctx0, cur, 0.5F);
-            }
-
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        cur = inpL;
-        struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-        cur = ggml_reshape_2d(ctx0, cur, n_embd, n_tokens);
-        cur = ggml_get_rows(ctx0, cur, inp_out_ids);
-
-        cur = llm_build_norm(ctx0, cur, hparams, model.output_norm, model.output_norm_b, LLM_NORM, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
-    // ref: https://github.com/facebookresearch/chameleon
-    // based on the original build_llama() function, changes:
-    //   * qk-norm
-    //   * swin-norm
-    //   * removed bias
-    //   * removed MoE
-    struct ggml_cgraph * build_chameleon() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, llama_model_max_nodes(model), false);
-
-        // mutable variable, needed during the last layer of the computation to skip unused tokens
-        int32_t n_tokens = this->n_tokens;
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head == hparams.n_rot);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * inpSA = inpL;
-
-            // norm
-            if (hparams.swin_norm) {
-                cur = inpL;
-            } else {
-                cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm, NULL,
-                    LLM_NORM_RMS, cb, il);
-                cb(cur, "attn_norm", il);
-            }
-
-            // self-attention
-            {
-                // compute Q and K and RoPE them
-                struct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
-                cb(Qcur, "Qcur", il);
-
-                struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
-                cb(Kcur, "Kcur", il);
-
-                struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
-                cb(Vcur, "Vcur", il);
-
-                if (model.layers[il].attn_q_norm) {
-                    Qcur = ggml_view_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens,
-                                ggml_element_size(Qcur) * n_embd_head,
-                                ggml_element_size(Qcur) * n_embd_head * n_head,
-                                0);
-                    cb(Qcur, "Qcur", il);
-
-                    Qcur = llm_build_norm(ctx0, Qcur, hparams,
-                                model.layers[il].attn_q_norm,
-                                model.layers[il].attn_q_norm_b,
-                                LLM_NORM, cb, il);
-                    cb(Qcur, "Qcur", il);
-                }
-
-                if (model.layers[il].attn_k_norm) {
-                    Kcur = ggml_view_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens,
-                                ggml_element_size(Kcur) * n_embd_head,
-                                ggml_element_size(Kcur) * n_embd_head * n_head_kv,
-                                0);
-                    cb(Kcur, "Kcur", il);
-
-                    Kcur = llm_build_norm(ctx0, Kcur, hparams,
-                               model.layers[il].attn_k_norm,
-                               model.layers[il].attn_k_norm_b,
-                               LLM_NORM, cb, il);
-                    cb(Kcur, "Kcur", il);
-                }
-
-                Qcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Qcur, "Qcur", il);
-
-                Kcur = ggml_rope_ext(
-                    ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, nullptr,
-                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
-                    ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(Kcur, "Kcur", il);
-
-                cur = llm_build_kv(ctx0, lctx, kv_self, gf,
-                        model.layers[il].wo, nullptr,
-                        Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-
-                if (hparams.swin_norm) {
-                    cur = llm_build_norm(ctx0, cur, hparams,
-                        model.layers[il].attn_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                }
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                n_tokens = n_outputs;
-                cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);
-                inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
-            }
-
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            if (!hparams.swin_norm) {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "ffn_norm", il);
-            }
-
-            cur = llm_build_ffn(ctx0, lctx, cur,
-                    model.layers[il].ffn_up,   NULL, NULL,
-                    model.layers[il].ffn_gate, NULL, NULL,
-                    model.layers[il].ffn_down, NULL, NULL,
-                    NULL,
-                    LLM_FFN_SILU, LLM_FFN_PAR, cb, il);
-            cb(cur, "ffn_out", il);
-
-            if (hparams.swin_norm) {
-                cur = llm_build_norm(ctx0, cur, hparams,
-                        model.layers[il].ffn_norm, NULL,
-                        LLM_NORM_RMS, cb, il);
-                cb(cur, "ffn_norm", il);
-            }
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "ffn_out", il);
-
-            cur = lctx.cvec.apply_to(ctx0, cur, il);
-            cb(cur, "l_out", il);
-
-            // input for next layer
-            inpL = cur;
-        }
-
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm, NULL,
-                LLM_NORM_RMS, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        // lm_head
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
-        cb(cur, "result_output_with_img_logits", -1);
-
-        // TODO: this suppresses the output of image tokens, which is required to enable text-only outputs.
-        // Needs to be removed once image outputs are supported.
-        int img_token_end_idx = 8196;
-        int img_token_start_idx = 4;
-        int num_img_tokens = img_token_end_idx - img_token_start_idx;
-        // creates 1d tensor of size num_img_tokens and values -FLT_MAX,
-        // which ensures that text token values are always at least larger than image token values
-        struct ggml_tensor * img_logits = ggml_new_tensor_1d(ctx0, GGML_TYPE_F32, num_img_tokens);
-        img_logits = ggml_clamp(ctx0, img_logits, -FLT_MAX, -FLT_MAX);
-        cb(img_logits, "img_logits", -1);
-        cur = ggml_set_1d(ctx0, cur, img_logits, ggml_element_size(cur) * img_token_start_idx);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-};
-
-static struct ggml_cgraph * llama_build_graph_defrag(llama_context & lctx, const std::vector & ids) {
-    llama_ubatch dummy = {};
-    dummy.equal_seqs = true;
-
-    llm_build_cb cb = [&](struct ggml_tensor * , const char * , int ) { };
-
-    struct llm_build_context llm(lctx, dummy, cb, false);
-
-    llm.init();
-
-    struct ggml_cgraph * result = llm.build_defrag(ids);
-
-    llm.free();
-
-    return result;
-}
-
-static struct ggml_cgraph * llama_build_graph_k_shift(llama_context & lctx) {
-    llama_ubatch dummy = {};
-    dummy.equal_seqs = true;
-
-    llm_build_cb cb = [&](struct ggml_tensor * , const char * , int ) { };
-
-    struct llm_build_context llm(lctx, dummy, cb, false);
-
-    llm.init();
-
-    struct ggml_cgraph * result = llm.build_k_shift();
-
-    llm.free();
-
-    return result;
-}
-
-static struct ggml_cgraph * llama_build_graph(
-         llama_context & lctx,
-    const llama_ubatch & ubatch,
-                  bool   worst_case) {
-    const auto & model = lctx.model;
-
-    // this callback allows us to apply custom logic to each tensor (e.g. ggml-alloc, offloading, etc.)
-    llm_build_cb cb = [&](struct ggml_tensor * cur, const char * name, int il) {
-        if (il >= 0) {
-            ggml_format_name(cur, "%s-%d", name, il);
-        } else {
-            ggml_set_name(cur, name);
-        }
-
-        if (!lctx.cparams.offload_kqv) {
-            if (strcmp(name, "kqv_merged_cont") == 0) {
-                // all nodes between the KV store and the attention output are run on the CPU
-                ggml_backend_sched_set_tensor_backend(lctx.sched.get(), cur, lctx.backend_cpu);
-            }
-        }
-
-        // norm may be automatically assigned to the backend of the previous layer, increasing data transfer between backends
-        // FIXME: fix in ggml_backend_sched
-        const bool full_offload = lctx.model.n_gpu_layers > (int)lctx.model.hparams.n_layer;
-        if (ubatch.n_tokens < 32 || full_offload) {
-            if (il != -1 && strcmp(name, "norm") == 0) {
-                const auto & dev_layer = lctx.model.dev_layer.at(il);
-                for (auto & backend : lctx.backends) {
-                    if (ggml_backend_get_device(backend.get()) == dev_layer.dev) {
-                        if (ggml_backend_supports_op(backend.get(), cur)) {
-                            ggml_backend_sched_set_tensor_backend(lctx.sched.get(), cur, backend.get());
-                        }
-                    }
-                }
-            }
-        }
-    };
-
-    struct ggml_cgraph * result = NULL;
-
-    struct llm_build_context llm(lctx, ubatch, cb, worst_case);
-
-    llm.init();
-
-    switch (model.arch) {
-        case LLM_ARCH_LLAMA:
-        case LLM_ARCH_MINICPM:
-        case LLM_ARCH_GRANITE:
-        case LLM_ARCH_GRANITE_MOE:
-            {
-                result = llm.build_llama();
-            } break;
-        case LLM_ARCH_BAICHUAN:
-            {
-                result = llm.build_baichuan();
-            } break;
-        case LLM_ARCH_FALCON:
-            {
-                result = llm.build_falcon();
-            } break;
-        case LLM_ARCH_GROK:
-            {
-                result = llm.build_grok();
-            } break;
-        case LLM_ARCH_STARCODER:
-            {
-                result = llm.build_starcoder();
-            } break;
-        case LLM_ARCH_REFACT:
-            {
-                result = llm.build_refact();
-            } break;
-        case LLM_ARCH_BERT:
-        case LLM_ARCH_JINA_BERT_V2:
-        case LLM_ARCH_NOMIC_BERT:
-            {
-                result = llm.build_bert();
-            } break;
-        case LLM_ARCH_BLOOM:
-            {
-                result = llm.build_bloom();
-            } break;
-        case LLM_ARCH_MPT:
-            {
-                result = llm.build_mpt();
-            } break;
-         case LLM_ARCH_STABLELM:
-            {
-                result = llm.build_stablelm();
-            } break;
-        case LLM_ARCH_QWEN:
-            {
-                result = llm.build_qwen();
-            } break;
-        case LLM_ARCH_QWEN2:
-            {
-                result = llm.build_qwen2();
-            } break;
-        case LLM_ARCH_QWEN2VL:
-            {
-                lctx.n_pos_per_token = 4;
-                result = llm.build_qwen2vl();
-            } break;
-        case LLM_ARCH_QWEN2MOE:
-            {
-                result = llm.build_qwen2moe();
-            } break;
-        case LLM_ARCH_PHI2:
-            {
-                result = llm.build_phi2();
-            } break;
-        case LLM_ARCH_PHI3:
-            {
-                result = llm.build_phi3();
-            } break;
-        case LLM_ARCH_PLAMO:
-            {
-                result = llm.build_plamo();
-            } break;
-        case LLM_ARCH_GPT2:
-            {
-                result = llm.build_gpt2();
-            } break;
-        case LLM_ARCH_CODESHELL:
-            {
-                result = llm.build_codeshell();
-            } break;
-        case LLM_ARCH_ORION:
-            {
-                result = llm.build_orion();
-            } break;
-        case LLM_ARCH_INTERNLM2:
-            {
-                result = llm.build_internlm2();
-            } break;
-        case LLM_ARCH_MINICPM3:
-            {
-                result = llm.build_minicpm3();
-            } break;
-        case LLM_ARCH_GEMMA:
-            {
-                result = llm.build_gemma();
-            } break;
-        case LLM_ARCH_GEMMA2:
-            {
-                result = llm.build_gemma2();
-            } break;
-        case LLM_ARCH_STARCODER2:
-            {
-                result = llm.build_starcoder2();
-            } break;
-        case LLM_ARCH_MAMBA:
-            {
-                result = llm.build_mamba();
-            } break;
-        case LLM_ARCH_XVERSE:
-            {
-                result = llm.build_xverse();
-            } break;
-        case LLM_ARCH_COMMAND_R:
-            {
-                result = llm.build_command_r();
-            } break;
-        case LLM_ARCH_DBRX:
-            {
-                result = llm.build_dbrx();
-            } break;
-        case LLM_ARCH_OLMO:
-            {
-                result = llm.build_olmo();
-            } break;
-        case LLM_ARCH_OLMO2:
-            {
-                result = llm.build_olmo2();
-            } break;
-        case LLM_ARCH_OLMOE:
-            {
-                result = llm.build_olmoe();
-            } break;
-        case LLM_ARCH_OPENELM:
-            {
-                result = llm.build_openelm();
-            } break;
-        case LLM_ARCH_GPTNEOX:
-            {
-                result = llm.build_gptneox();
-            } break;
-        case LLM_ARCH_ARCTIC:
-            {
-                result = llm.build_arctic();
-            } break;
-        case LLM_ARCH_DEEPSEEK:
-            {
-                result = llm.build_deepseek();
-            } break;
-        case LLM_ARCH_DEEPSEEK2:
-            {
-                result = llm.build_deepseek2();
-            } break;
-        case LLM_ARCH_CHATGLM:
-            {
-                result = llm.build_chatglm();
-            } break;
-        case LLM_ARCH_BITNET:
-            {
-                result = llm.build_bitnet();
-            } break;
-        case LLM_ARCH_T5:
-            {
-                if (lctx.is_encoding) {
-                    result = llm.build_t5_encoder();
-                } else {
-                    result = llm.build_t5_decoder();
-                }
-            } break;
-        case LLM_ARCH_T5ENCODER:
-            {
-                result = llm.build_t5_encoder();
-            } break;
-        case LLM_ARCH_JAIS:
-            {
-                result = llm.build_jais();
-            } break;
-        case LLM_ARCH_NEMOTRON:
-            {
-                result = llm.build_nemotron();
-            } break;
-        case LLM_ARCH_EXAONE:
-            {
-                result = llm.build_exaone();
-            } break;
-        case LLM_ARCH_RWKV6:
-            {
-                result = llm.build_rwkv6();
-            } break;
-        case LLM_ARCH_CHAMELEON:
-            {
-                result = llm.build_chameleon();
-            } break;
-        default:
-            GGML_ABORT("fatal error");
-    }
-
-    // add on pooling layer
-    if (lctx.cparams.embeddings) {
-        result = llm.append_pooling(result);
-    }
-
-    llm.free();
-
-    return result;
-}
-
-static void llama_set_k_shift(llama_context & lctx) {
-    const int64_t kv_size = lctx.kv_self.size;
-
-    assert(ggml_backend_buffer_is_host(lctx.inp_K_shift->buffer));
-
-    int32_t * data = (int32_t *) lctx.inp_K_shift->data;
-
-    for (int i = 0; i < kv_size; ++i) {
-        data[i] = lctx.kv_self.cells[i].delta;
-    }
-}
-
-static void llama_set_s_copy(llama_context & lctx) {
-    const int64_t kv_size = lctx.kv_self.size;
-
-    assert(ggml_backend_buffer_is_host(lctx.inp_s_copy->buffer));
-
-    int32_t * data = (int32_t *) lctx.inp_s_copy->data;
-
-    for (int i = 0; i < kv_size; ++i) {
-        data[i] = lctx.kv_self.cells[i].src;
-    }
-}
-
-static int32_t llama_relative_position_bucket(llama_pos x, llama_pos y, uint64_t n_buckets, bool bidirectional) {
-    // TODO move to hparams if a T5 variant appears that uses a different value
-    const int64_t max_distance = 128;
-
-    if (bidirectional) {
-        n_buckets >>= 1;
-    }
-
-    const int64_t max_exact = n_buckets >> 1;
-
-    int32_t relative_position = x - y;
-    int32_t relative_bucket = 0;
-    if (bidirectional) {
-        relative_bucket += (relative_position > 0) * n_buckets;
-        relative_position = abs(relative_position);
-    } else {
-        relative_position = -std::min(relative_position, 0);
-    }
-    int32_t relative_position_if_large = floorf(max_exact + logf(1.0 * relative_position / max_exact) * (n_buckets - max_exact) / log(1.0 * max_distance / max_exact));
-    relative_position_if_large = std::min(relative_position_if_large, n_buckets - 1);
-    relative_bucket += (relative_position < max_exact ? relative_position : relative_position_if_large);
-    return relative_bucket;
-}
-
-static void llama_set_inputs(llama_context & lctx, const llama_ubatch & ubatch) {
-    //
-    // set input data
-    //
-
-    const auto & hparams = lctx.model.hparams;
-    const auto & cparams = lctx.cparams;
-    const auto & kv_self = lctx.kv_self;
-
-    if (ubatch.token) {
-        const int64_t n_tokens = ubatch.n_tokens;
-
-        ggml_backend_tensor_set(lctx.inp_tokens, ubatch.token, 0, n_tokens*ggml_element_size(lctx.inp_tokens));
-    }
-
-    if (ubatch.embd) {
-        const int64_t n_embd   = hparams.n_embd;
-        const int64_t n_tokens = ubatch.n_tokens;
-
-        ggml_backend_tensor_set(lctx.inp_embd, ubatch.embd, 0, n_tokens*n_embd*ggml_element_size(lctx.inp_embd));
-    }
-
-    if (ubatch.pos && lctx.inp_pos) {
-        const int64_t n_tokens = ubatch.n_tokens;
-        auto n_pos = lctx.n_pos_per_token;
-        ggml_backend_tensor_set(lctx.inp_pos, ubatch.pos, 0, n_tokens*n_pos*ggml_element_size(lctx.inp_pos));
-    }
-
-    if (hparams.causal_attn || cparams.pooling_type == LLAMA_POOLING_TYPE_NONE) {
-        GGML_ASSERT(lctx.inp_out_ids && "every model that can must skip unused outputs");
-        const int64_t n_tokens = ubatch.n_tokens;
-
-        GGML_ASSERT(ggml_backend_buffer_is_host(lctx.inp_out_ids->buffer));
-        int32_t * data = (int32_t *) lctx.inp_out_ids->data;
-
-        if (lctx.n_outputs == n_tokens) {
-            for (int i = 0; i < n_tokens; ++i) {
-                data[i] = i;
-            }
-        } else if (ubatch.output) {
-            int32_t n_outputs = 0;
-            for (int i = 0; i < n_tokens; ++i) {
-                if (ubatch.output[i]) {
-                    data[n_outputs++] = i;
-                }
-            }
-            // the graph needs to have been passed the correct number of outputs
-            GGML_ASSERT(lctx.n_outputs == n_outputs);
-        } else if (lctx.n_outputs == 1) {
-            // only keep last output
-            data[0] = n_tokens - 1;
-        } else {
-            GGML_ASSERT(lctx.n_outputs == 0);
-        }
-    }
-
-    GGML_ASSERT(
-        // (!a || b) is a logical implication (a -> b)
-        // !hparams.causal_attn -> !cparams.causal_attn
-        (hparams.causal_attn || !cparams.causal_attn) &&
-        "causal attention is not supported by this model"
-    );
-
-    if (lctx.inp_KQ_mask || lctx.inp_KQ_mask_swa) {
-        // NOTE: hparams.causal_attn indicates the model is capable of generation and uses the kv cache.
-        if (cparams.causal_attn && !lctx.is_encoding) {
-            const int64_t n_kv         = kv_self.n;
-            const int64_t n_tokens     = ubatch.n_tokens;
-            const int64_t n_seq_tokens = ubatch.n_seq_tokens;
-            const int64_t n_seqs       = ubatch.n_seqs;
-
-
-            float * data     = nullptr;
-            float * data_swa = nullptr;
-
-            if (lctx.inp_KQ_mask) {
-                GGML_ASSERT(ggml_backend_buffer_is_host(lctx.inp_KQ_mask->buffer));
-                data = (float *) lctx.inp_KQ_mask->data;
-            }
-
-            if (lctx.inp_KQ_mask_swa) {
-                GGML_ASSERT(ggml_backend_buffer_is_host(lctx.inp_KQ_mask_swa->buffer));
-                data_swa = (float *) lctx.inp_KQ_mask_swa->data;
-            }
-
-            // For causal attention, use only the previous KV cells
-            // of the correct sequence for each token of the ubatch.
-            // It's assumed that if a token in the batch has multiple sequences, they are equivalent.
-            for (int h = 0; h < 1; ++h) {
-                for (int s = 0; s < n_seqs; ++s) {
-                    const llama_seq_id seq_id = ubatch.seq_id[s][0];
-
-                    for (int j = 0; j < n_seq_tokens; ++j) {
-                        const llama_pos pos = ubatch.pos[s*n_seq_tokens + j];
-
-                        for (int i = 0; i < n_kv; ++i) {
-                            float f;
-                            if (!kv_self.cells[i].has_seq_id(seq_id) || kv_self.cells[i].pos > pos) {
-                                f = -INFINITY;
-                            } else {
-                                if (hparams.use_alibi) {
-                                    f = -std::abs(kv_self.cells[i].pos - pos);
-                                } else {
-                                    f = 0.0f;
-                                }
-                            }
-
-                            if (data) {
-                                data[h*(n_kv*n_tokens) + s*(n_kv*n_seq_tokens) + j*n_kv + i] = f;
-                            }
-
-                            // may need to cut off old tokens for sliding window
-                            if (data_swa) {
-                                if (pos - kv_self.cells[i].pos >= (int32_t)hparams.n_swa) {
-                                    f = -INFINITY;
-                                }
-                                data_swa[h*(n_kv*n_tokens) + s*(n_kv*n_seq_tokens) + j*n_kv + i] = f;
-                            }
-                        }
-                    }
-                }
-
-                if (data) {
-                    for (int i = n_tokens; i < GGML_PAD(n_tokens, GGML_KQ_MASK_PAD); ++i) {
-                        for (int j = 0; j < n_kv; ++j) {
-                            data[h*(n_kv*n_tokens) + i*n_kv + j] = -INFINITY;
-                        }
-                    }
-                }
-
-                if (data_swa) {
-                    for (int i = n_tokens; i < GGML_PAD(n_tokens, GGML_KQ_MASK_PAD); ++i) {
-                        for (int j = 0; j < n_kv; ++j) {
-                            data_swa[h*(n_kv*n_tokens) + i*n_kv + j] = -INFINITY;
-                        }
-                    }
-                }
-            }
-        } else {
-            const int64_t n_tokens     = ubatch.n_tokens;
-            const int64_t n_seq_tokens = ubatch.n_seq_tokens;
-            const int64_t n_seqs       = ubatch.n_seqs;
-            // when using kv cache, the mask needs to match the kv cache size
-            const int64_t n_stride = hparams.causal_attn && !lctx.is_encoding ? kv_self.n : n_tokens;
-
-            GGML_ASSERT(ggml_backend_buffer_is_host(lctx.inp_KQ_mask->buffer));
-
-            float * data = (float *) lctx.inp_KQ_mask->data;
-
-            for (int h = 0; h < 1; ++h) {
-                for (int s1 = 0; s1 < n_seqs; ++s1) {
-                    const llama_seq_id seq_id = ubatch.seq_id[s1][0];
-
-                    for (int j = 0; j < n_seq_tokens; ++j) {
-                        const int32_t tj = s1*n_seq_tokens + j;
-
-                        for (int s0 = 0; s0 < n_seqs; ++s0) {
-                            for (int i = 0; i < n_seq_tokens; ++i) {
-                                const int32_t ti = s0*n_seq_tokens + i;
-                                float f = -INFINITY;
-
-                                for (int s = 0; s < ubatch.n_seq_id[s0]; ++s) {
-                                    if (ubatch.seq_id[s0][s] == seq_id) {
-                                        if (hparams.use_alibi) {
-                                            f = -std::abs(ubatch.pos[ti] - ubatch.pos[tj]);
-                                        } else {
-                                            f = 0.0f;
-                                        }
-                                        break;
-                                    }
-                                }
-
-                                data[h*(n_tokens*n_tokens) + tj*n_stride + ti] = f;
-                            }
-                        }
-
-                        for (int i = n_tokens; i < n_stride; ++i) {
-                            data[h*(n_tokens*n_tokens) + tj*n_stride + i] = -INFINITY;
-                        }
-                    }
-                }
-            }
-        }
-    }
-
-    if (cparams.embeddings && cparams.pooling_type == LLAMA_POOLING_TYPE_MEAN) {
-        const int64_t n_tokens     = ubatch.n_tokens;
-        const int64_t n_seq_tokens = ubatch.n_seq_tokens;
-        const int64_t n_seqs       = ubatch.n_seqs;
-
-        GGML_ASSERT(lctx.inp_mean);
-        GGML_ASSERT(ggml_backend_buffer_is_host(lctx.inp_mean->buffer));
-
-        float * data = (float *) lctx.inp_mean->data;
-        memset(lctx.inp_mean->data, 0, n_tokens * n_tokens * ggml_element_size(lctx.inp_mean));
-
-        std::vector sum(n_tokens, 0);
-
-        for (int s = 0; s < n_seqs; ++s) {
-            const llama_seq_id seq_id = ubatch.seq_id[s][0];
-
-            // TODO: adapt limits to n_seqs when ubatch.equal_seqs is true
-            GGML_ASSERT(seq_id < n_tokens && "seq_id cannot be larger than n_tokens with pooling_type == MEAN");
-
-            sum[seq_id] += ubatch.n_seq_tokens;
-        }
-
-        std::vector div(n_tokens, 0.0f);
-        for (int i = 0; i < n_tokens; ++i) {
-            const uint64_t s = sum[i];
-            if (s > 0) {
-                div[i] = 1.0f/float(s);
-            }
-        }
-
-        for (int s = 0; s < n_seqs; ++s) {
-            const llama_seq_id seq_id = ubatch.seq_id[s][0];
-
-            for (int i = 0; i < n_seq_tokens; ++i) {
-                data[seq_id*n_tokens + s*n_seq_tokens + i] = div[seq_id];
-            }
-        }
-    }
-
-    if (cparams.embeddings && (
-                cparams.pooling_type == LLAMA_POOLING_TYPE_CLS ||
-                cparams.pooling_type == LLAMA_POOLING_TYPE_RANK)) {
-        const int64_t n_tokens     = ubatch.n_tokens;
-        const int64_t n_seq_tokens = ubatch.n_seq_tokens;
-        const int64_t n_seqs       = ubatch.n_seqs;
-
-        GGML_ASSERT(lctx.inp_cls);
-        GGML_ASSERT(ggml_backend_buffer_is_host(lctx.inp_cls->buffer));
-
-        uint32_t * data = (uint32_t *) lctx.inp_cls->data;
-        memset(lctx.inp_cls->data, 0, n_tokens * ggml_element_size(lctx.inp_cls));
-
-        for (int s = 0; s < n_seqs; ++s) {
-            const llama_seq_id seq_id = ubatch.seq_id[s][0];
-
-            // TODO: adapt limits to n_seqs when ubatch.equal_seqs is true
-            GGML_ASSERT(seq_id < n_tokens && "seq_id cannot be larger than n_tokens with pooling_type == CLS or RANK");
-
-            for (int i = 0; i < n_seq_tokens; ++i) {
-                const llama_pos pos = ubatch.pos[s*n_seq_tokens + i];
-
-                if (pos == 0) {
-                    data[seq_id] = s*n_seq_tokens + i;
-                }
-            }
-        }
-    }
-
-    if (cparams.embeddings && cparams.pooling_type == LLAMA_POOLING_TYPE_LAST) {
-        const int64_t n_tokens     = ubatch.n_tokens;
-        const int64_t n_seq_tokens = ubatch.n_seq_tokens;
-        const int64_t n_seqs       = ubatch.n_seqs;
-
-        GGML_ASSERT(lctx.inp_cls);
-        GGML_ASSERT(ggml_backend_buffer_is_host(lctx.inp_cls->buffer));
-
-        uint32_t * data = (uint32_t *) lctx.inp_cls->data;
-        memset(lctx.inp_cls->data, 0, n_tokens * ggml_element_size(lctx.inp_cls));
-
-        std::vector last_pos(n_tokens, -1);
-        std::vector last_row(n_tokens, -1);
-
-        for (int s = 0; s < n_seqs; ++s) {
-            const llama_seq_id seq_id = ubatch.seq_id[s][0];
-
-            // TODO: adapt limits to n_seqs when ubatch.equal_seqs is true
-            GGML_ASSERT(seq_id < n_tokens && "seq_id cannot be larger than n_tokens with pooling_type == LAST");
-
-            for (int i = 0; i < n_seq_tokens; ++i) {
-                const llama_pos pos = ubatch.pos[s*n_seq_tokens + i];
-
-                if (pos >= last_pos[seq_id]) {
-                    last_pos[seq_id] = pos;
-                    last_row[seq_id] = s*n_seq_tokens + i;
-                }
-            }
-        }
-
-        for (int i = 0; i < n_tokens; ++i) {
-            if (last_row[i] >= 0) {
-                data[i] = last_row[i];
-            }
-        }
-    }
-
-    if (kv_self.recurrent) {
-        const int64_t n_kv = kv_self.n;
-
-        if (lctx.inp_s_mask) {
-            GGML_ASSERT(ggml_backend_buffer_is_host(lctx.inp_s_mask->buffer));
-            float * data = (float *) lctx.inp_s_mask->data;
-
-            // clear unused states
-            for (int i = 0; i < n_kv; ++i) {
-                const uint32_t  cell_id = i + kv_self.head;
-                llama_kv_cell & kv_cell = lctx.kv_self.cells[cell_id];
-
-                data[i] = (float) (kv_cell.src >= 0);
-
-                // only clear once
-                if (kv_cell.src < 0) {
-                    kv_cell.src = cell_id;
-                }
-            }
-        }
-
-        if (lctx.inp_s_copy) {
-            GGML_ASSERT(ggml_backend_buffer_is_host(lctx.inp_s_copy->buffer));
-            int32_t * data = (int32_t *) lctx.inp_s_copy->data;
-
-            // assuming copy destinations ALWAYS happen ONLY on the cells between head and head+n
-            for (uint32_t i = 0; i < n_kv; ++i) {
-                const uint32_t  cell_id = i + kv_self.head;
-                llama_kv_cell & kv_cell = lctx.kv_self.cells[cell_id];
-
-                // prevent out-of-bound sources
-                if (kv_cell.src < 0 || (uint32_t) kv_cell.src >= kv_self.size) {
-                    kv_cell.src = cell_id;
-                }
-
-                data[i] = kv_cell.src;
-
-                // ensure copy only happens once
-                if (kv_cell.src != (int32_t) cell_id) {
-                    kv_cell.src = cell_id;
-                }
-            }
-        }
-    }
-
-    if (lctx.inp_pos_bucket) {
-        const int64_t n_tokens = ubatch.n_tokens;
-
-        GGML_ASSERT(ggml_backend_buffer_is_host(lctx.inp_pos_bucket->buffer));
-        GGML_ASSERT(!ubatch.equal_seqs); // TODO: use ubatch.n_seqs instead of failing
-
-        int32_t * data = (int32_t *) lctx.inp_pos_bucket->data;
-
-        if (!lctx.is_encoding) {
-            const int64_t n_kv = kv_self.n;
-            for (int h = 0; h < 1; ++h) {
-                for (int j = 0; j < n_tokens; ++j) {
-                    for (int i = 0; i < n_kv; ++i) {
-                        data[h*(n_kv*n_tokens) + j*n_kv + i] = llama_relative_position_bucket(lctx.kv_self.cells[i].pos, ubatch.pos[j], hparams.n_rel_attn_bkts, lctx.is_encoding);
-                    }
-                }
-            }
-        } else {
-            for (int h = 0; h < 1; ++h) {
-                for (int j = 0; j < n_tokens; ++j) {
-                    for (int i = 0; i < n_tokens; ++i) {
-                        data[h*(n_tokens*n_tokens) + j*n_tokens + i] = llama_relative_position_bucket(ubatch.pos[i], ubatch.pos[j], hparams.n_rel_attn_bkts, lctx.is_encoding);
-                    }
-                }
-            }
-        }
-    }
-
-    if (!lctx.is_encoding && lctx.inp_embd_enc) {
-        assert(lctx.inp_embd_enc->type == GGML_TYPE_F32);
-        assert((size_t) ggml_nelements(lctx.inp_embd_enc) == lctx.embd_enc.size());
-
-        ggml_backend_tensor_set(lctx.inp_embd_enc, lctx.embd_enc.data(), 0, ggml_nbytes(lctx.inp_embd_enc));
-    }
-
-    if (!lctx.is_encoding && lctx.inp_KQ_mask_cross) {
-        const int64_t n_output_enc = lctx.embd_enc.size() / hparams.n_embd;
-        const int64_t n_tokens = ubatch.n_tokens;
-
-        GGML_ASSERT(ggml_backend_buffer_is_host(lctx.inp_KQ_mask_cross->buffer));
-        GGML_ASSERT(!ubatch.equal_seqs); // TODO: use ubatch.n_seqs instead of failing
-
-        float * data = (float *) lctx.inp_KQ_mask_cross->data;
-
-        for (int h = 0; h < 1; ++h) {
-            for (int j = 0; j < n_tokens; ++j) {
-                for (int i = 0; i < n_output_enc; ++i) {
-                    float f = -INFINITY;
-                    for (int s = 0; s < ubatch.n_seq_id[j]; ++s) {
-                        const llama_seq_id seq_id = ubatch.seq_id[j][s];
-                        if (lctx.seq_ids_enc[i].find(seq_id) != lctx.seq_ids_enc[i].end()) {
-                            f = 0.0f;
-                        }
-                    }
-                    data[h*(n_output_enc*n_tokens) + j*n_output_enc + i] = f;
-                }
-            }
-
-            for (int i = n_tokens; i < GGML_PAD(n_tokens, GGML_KQ_MASK_PAD); ++i) {
-                for (int j = 0; j < n_output_enc; ++j) {
-                    data[h*(n_output_enc*n_tokens) + i*n_output_enc + j] = -INFINITY;
-                }
-            }
-        }
-    }
-}
-
-// Make sure enough space is available for outputs.
-// Returns max number of outputs for which space was reserved.
-static size_t llama_output_reserve(llama_context & lctx, size_t n_outputs) {
-    const auto & cparams = lctx.cparams;
-    const auto & hparams = lctx.model.hparams;
-
-    const size_t n_outputs_max = std::max(n_outputs, (size_t) cparams.n_seq_max);
-
-    const auto n_batch = cparams.n_batch;
-    const auto n_vocab = hparams.n_vocab;
-    const auto n_embd  = hparams.n_embd;
-
-    // TODO: use a per-batch flag for logits presence instead
-    const bool has_logits = !cparams.embeddings;
-    const bool has_embd   =  cparams.embeddings && (cparams.pooling_type == LLAMA_POOLING_TYPE_NONE);
-
-    const size_t logits_size = has_logits ? n_vocab*n_outputs_max : 0;
-    const size_t embd_size   = has_embd   ?  n_embd*n_outputs_max : 0;
-
-    if (lctx.output_ids.empty()) {
-        // init, never resized afterwards
-        lctx.output_ids.resize(n_batch);
-    }
-
-    const size_t prev_size = lctx.buf_output ? ggml_backend_buffer_get_size(lctx.buf_output.get()) : 0;
-    const size_t new_size  = (logits_size + embd_size) * sizeof(float);
-
-    // alloc only when more than the current capacity is required
-    // TODO: also consider shrinking the buffer
-    if (!lctx.buf_output || prev_size < new_size) {
-        if (lctx.buf_output) {
-#ifndef NDEBUG
-            // This doesn't happen often, but may be annoying in some cases (like the HellaSwag benchmark)
-            LLAMA_LOG_INFO("%s: reallocating output buffer from size %.02f MiB to %.02f MiB\n", __func__, prev_size / 1024.0 / 1024.0, new_size / 1024.0 / 1024.0);
-#endif
-            lctx.buf_output = nullptr;
-            lctx.logits = nullptr;
-            lctx.embd = nullptr;
-        }
-
-        auto * buft = ggml_backend_cpu_buffer_type();
-        // try to use the host buffer of the device where the output tensor is allocated for faster transfer to system memory
-        auto * output_dev = lctx.model.dev_output.dev;
-        auto * output_dev_host_buft = output_dev ? ggml_backend_dev_host_buffer_type(output_dev) : nullptr;
-        if (output_dev_host_buft) {
-            buft = output_dev_host_buft;
-        }
-        lctx.buf_output.reset(ggml_backend_buft_alloc_buffer(buft, new_size));
-        if (lctx.buf_output == nullptr) {
-            LLAMA_LOG_ERROR("%s: failed to allocate output buffer of size %.2f MiB\n", __func__, new_size / (1024.0 * 1024.0));
-            return 0;
-        }
-    }
-
-    float * output_base = (float *) ggml_backend_buffer_get_base(lctx.buf_output.get());
-
-    lctx.logits = has_logits ? output_base               : nullptr;
-    lctx.embd   = has_embd   ? output_base + logits_size : nullptr;
-
-    lctx.output_size = n_outputs_max;
-    lctx.logits_size = logits_size;
-    lctx.embd_size   = embd_size;
-
-    // set all ids as invalid (negative)
-    std::fill(lctx.output_ids.begin(), lctx.output_ids.end(), -1);
-
-    ggml_backend_buffer_clear(lctx.buf_output.get(), 0);
-
-    lctx.n_outputs = 0;
-
-    return n_outputs_max;
-}
-
-// make the outputs have the same order they had in the user-provided batch
-static void llama_output_reorder(struct llama_context * ctx) {
-    std::vector & out_ids = ctx->sbatch.out_ids;
-    if (!out_ids.empty()) {
-        uint32_t n_vocab = ctx->model.hparams.n_vocab;
-        uint32_t n_embd  = ctx->model.hparams.n_embd;
-        int32_t n_outputs = ctx->n_outputs;
-        GGML_ASSERT((size_t) n_outputs == out_ids.size());
-        // TODO: is there something more efficient which also minimizes swaps?
-        // selection sort, to minimize swaps (from https://en.wikipedia.org/wiki/Selection_sort)
-        for (int32_t i = 0; i < n_outputs - 1; ++i) {
-            int32_t j_min = i;
-            for (int32_t j = i + 1; j < n_outputs; ++j) {
-                if (out_ids[j] < out_ids[j_min]) {
-                    j_min = j;
-                }
-            }
-            if (j_min == i) { continue; }
-            std::swap(out_ids[i], out_ids[j_min]);
-            if (ctx->logits_size > 0) {
-                for (uint32_t k = 0; k < n_vocab; k++) {
-                    std::swap(ctx->logits[i*n_vocab + k], ctx->logits[j_min*n_vocab + k]);
-                }
-            }
-            if (ctx->embd_size > 0) {
-                for (uint32_t k = 0; k < n_embd; k++) {
-                    std::swap(ctx->embd[i*n_embd + k], ctx->embd[j_min*n_embd + k]);
-                }
-            }
-        }
-        std::fill(ctx->output_ids.begin(), ctx->output_ids.end(), -1);
-        for (int32_t i = 0; i < n_outputs; ++i) {
-            ctx->output_ids[out_ids[i]] = i;
-        }
-        out_ids.clear();
-    }
-}
-
-// returns the result of ggml_backend_sched_graph_compute_async execution
-static enum ggml_status llama_graph_compute(
-          llama_context & lctx,
-            ggml_cgraph * gf,
-                    int   n_threads,
-        ggml_threadpool * threadpool) {
-    if (lctx.backend_cpu != nullptr) {
-        auto * reg = ggml_backend_dev_backend_reg(ggml_backend_get_device(lctx.backend_cpu));
-        auto * set_threadpool_fn = (decltype(ggml_backend_cpu_set_threadpool) *) ggml_backend_reg_get_proc_address(reg, "ggml_backend_cpu_set_threadpool");
-        set_threadpool_fn(lctx.backend_cpu, threadpool);
-    }
-
-    // set the number of threads for all the backends
-    for (const auto & set_n_threads_fn : lctx.set_n_threads_fns) {
-        set_n_threads_fn.second(set_n_threads_fn.first, n_threads);
-    }
-
-    auto status = ggml_backend_sched_graph_compute_async(lctx.sched.get(), gf);
-    if (status != GGML_STATUS_SUCCESS) {
-        LLAMA_LOG_ERROR("%s: ggml_backend_sched_graph_compute_async failed with error %d\n", __func__, status);
-    }
-
-    // fprintf(stderr, "splits: %d\n", ggml_backend_sched_get_n_splits(lctx.sched));
-
-    return status;
-}
-
-// decode a batch of tokens by evaluating the transformer
-// in case of unsuccessful decoding (error or warning),
-// the kv_cache state will be returned to its original state
-// (for non-recurrent models) or cleaned (for recurrent models)
-//
-//   - lctx:      llama context
-//   - batch:     batch to evaluate
-//
-// return 0 on success
-// return positive int on warning
-// return negative int on error
-//
-static int llama_decode_internal(
-         llama_context & lctx,
-           llama_batch   inp_batch) {
-
-    lctx.is_encoding = false;
-
-    if (inp_batch.n_tokens == 0) {
-        LLAMA_LOG_ERROR("%s: n_tokens == 0\n", __func__);
-        return -1;
-    }
-
-    // temporary allocate memory for the input batch if needed
-    llama_batch_allocr batch_allocr(lctx, inp_batch);
-    const llama_batch & batch = batch_allocr.batch;
-    const uint32_t n_tokens_all = batch.n_tokens;
-
-    const auto & model   = lctx.model;
-    const auto & hparams = model.hparams;
-    const auto & cparams = lctx.cparams;
-
-    GGML_ASSERT((!batch.token && batch.embd) || (batch.token && !batch.embd)); // NOLINT
-
-    if (batch.token) {
-        for (uint32_t i = 0; i < n_tokens_all; ++i) {
-            if (batch.token[i] < 0 || (uint32_t)batch.token[i] >= model.vocab.n_vocab) {
-                LLAMA_LOG_ERROR("%s: invalid token[%d] = %d\n", __func__, i, batch.token[i]);
-                return -1;
-            }
-        }
-    }
-
-    GGML_ASSERT(n_tokens_all <= cparams.n_batch);
-
-    GGML_ASSERT((cparams.causal_attn || cparams.n_ubatch >= n_tokens_all) && "non-causal attention requires n_ubatch >= n_tokens");
-
-    if (lctx.t_compute_start_us == 0) {
-        lctx.t_compute_start_us = ggml_time_us();
-    }
-    lctx.n_queued_tokens += n_tokens_all;
-
-    auto & kv_self = lctx.kv_self;
-    llama_kv_slot_restorer kv_slot_restorer(kv_self);
-
-    const int64_t n_embd  = hparams.n_embd;
-    const int64_t n_vocab = hparams.n_vocab;
-
-    uint32_t n_outputs = 0;
-    uint32_t n_outputs_prev = 0;
-
-    const auto n_ubatch = cparams.n_ubatch;
-
-    // this indicates we are doing pooled embedding, so we ignore batch.logits and output all tokens
-    const bool embd_pooled = cparams.embeddings && cparams.pooling_type != LLAMA_POOLING_TYPE_NONE;
-
-    lctx.embd_seq.clear();
-
-    // count outputs
-    if (batch.logits && !embd_pooled) {
-        for (uint32_t i = 0; i < n_tokens_all; ++i) {
-            n_outputs += batch.logits[i] != 0;
-        }
-    } else if (lctx.logits_all || embd_pooled) {
-        n_outputs = n_tokens_all;
-    } else {
-        // keep last output only
-        n_outputs = 1;
-    }
-
-    lctx.sbatch.from_batch(batch, n_embd,
-        /* simple_split */ !kv_self.recurrent,
-        /* logits_all   */ n_outputs == n_tokens_all);
-
-    // reserve output buffer
-    if (llama_output_reserve(lctx, n_outputs) < n_outputs) {
-        LLAMA_LOG_ERROR("%s: could not reserve space for batch with %u outputs\n", __func__, n_outputs);
-        return -2;
-    };
-
-    while (lctx.sbatch.n_tokens > 0) {
-        llama_ubatch ubatch;
-        if (kv_self.recurrent) {
-            if (embd_pooled) {
-                // Pooled embeddings cannot be split across ubatches (yet)
-                ubatch = lctx.sbatch.split_seq(n_ubatch);
-            } else {
-                // recurrent model architectures are easier to implement
-                // with equal-length sequences
-                ubatch = lctx.sbatch.split_equal(n_ubatch);
-            }
-        } else {
-            ubatch = lctx.sbatch.split_simple(n_ubatch);
-        }
-        const uint32_t n_tokens = ubatch.n_tokens;
-
-        // count the outputs in this u_batch
-        {
-            int32_t n_outputs_new = 0;
-
-            if (n_outputs == n_tokens_all) {
-                n_outputs_new = n_tokens;
-            } else {
-                GGML_ASSERT(ubatch.output);
-                for (uint32_t i = 0; i < n_tokens; i++) {
-                    n_outputs_new += (int32_t) (ubatch.output[i] != 0);
-                }
-            }
-
-            // needs to happen before the graph is built
-            lctx.n_outputs = n_outputs_new;
-        }
-
-        int n_threads = n_tokens == 1 ? cparams.n_threads : cparams.n_threads_batch;
-        ggml_threadpool_t threadpool = n_tokens == 1 ? lctx.threadpool : lctx.threadpool_batch;
-
-        GGML_ASSERT(n_threads > 0);
-
-        // non-causal masks do not use the KV cache
-        if (hparams.causal_attn) {
-            llama_kv_cache_update(&lctx);
-
-            // if we have enough unused cells before the current head ->
-            //   better to start searching from the beginning of the cache, hoping to fill it
-            if (kv_self.head > kv_self.used + 2*n_tokens) {
-                kv_self.head = 0;
-            }
-
-            const auto slot = llama_kv_cache_find_slot(kv_self, ubatch);
-            if (!slot) {
-                return 1;
-            }
-            kv_slot_restorer.save(slot);
-
-            if (!kv_self.recurrent) {
-                // a heuristic, to avoid attending the full cache if it is not yet utilized
-                // after enough generations, the benefit from this heuristic disappears
-                // if we start defragmenting the cache, the benefit from this will be more important
-                const uint32_t pad = llama_kv_cache_get_padding(cparams);
-                kv_self.n = std::min(kv_self.size, std::max(pad, GGML_PAD(llama_kv_cache_cell_max(kv_self), pad)));
-                //kv_self.n = llama_kv_cache_cell_max(kv_self);
-            }
-        }
-
-        //printf("kv_self.n = %5d, kv_self.used = %5d, kv_self.head = %5d\n", kv_self.n, kv_self.used, kv_self.head);
-
-        ggml_backend_sched_reset(lctx.sched.get());
-        ggml_backend_sched_set_eval_callback(lctx.sched.get(), lctx.cparams.cb_eval, lctx.cparams.cb_eval_user_data);
-
-        ggml_cgraph * gf = llama_build_graph(lctx, ubatch, false);
-
-        // the output is always the last tensor in the graph
-        struct ggml_tensor * res  = ggml_graph_node(gf, -1);
-        struct ggml_tensor * embd = ggml_graph_node(gf, -2);
-
-        if (lctx.n_outputs == 0) {
-            // no output
-            res  = nullptr;
-            embd = nullptr;
-        } else if (cparams.embeddings) {
-            res  = nullptr; // do not extract logits for embedding case
-            embd = nullptr;
-            for (int i = ggml_graph_n_nodes(gf) - 1; i >= 0; --i) {
-                if (strcmp(ggml_graph_node(gf, i)->name, "result_embd_pooled") == 0) {
-                    embd = ggml_graph_node(gf, i);
-                    break;
-                }
-            }
-            GGML_ASSERT(embd != nullptr && "missing embeddings tensor");
-        } else {
-            embd = nullptr; // do not extract embeddings when not needed
-            GGML_ASSERT(strcmp(res->name, "result_output") == 0 && "missing result_output tensor");
-        }
-        // LLAMA_LOG_INFO("graph build time: %.3f ms (%d nodes, %d leafs)\n", (ggml_time_us() - t_start_us)/1000.0, gf->n_nodes, gf->n_leafs);
-
-        ggml_backend_sched_alloc_graph(lctx.sched.get(), gf);
-
-        llama_set_inputs(lctx, ubatch);
-
-        const auto compute_status = llama_graph_compute(lctx, gf, n_threads, threadpool);
-        if (compute_status != GGML_STATUS_SUCCESS) {
-            kv_slot_restorer.restore(kv_self);
-            switch (compute_status) {
-                case GGML_STATUS_ABORTED:
-                    return 2;
-                case GGML_STATUS_ALLOC_FAILED:
-                    return -2;
-                case GGML_STATUS_FAILED:
-                default:
-                    return -3;
-            }
-        }
-
-        // update the kv ring buffer
-        {
-            kv_self.head += n_tokens;
-
-            // Ensure kv cache head points to a valid index.
-            if (kv_self.head >= kv_self.size) {
-                kv_self.head = 0;
-            }
-        }
-
-        // plot the computation graph in dot format (for debugging purposes)
-        //if (n_past%100 == 0) {
-        //    ggml_graph_dump_dot(gf, NULL, "llama.dot");
-        //}
-
-        // extract logits
-        if (res) {
-            ggml_backend_t backend_res = ggml_backend_sched_get_tensor_backend(lctx.sched.get(), res);
-            GGML_ASSERT(backend_res != nullptr);
-            GGML_ASSERT(lctx.logits != nullptr);
-
-            float * logits_out = lctx.logits + n_outputs_prev*n_vocab;
-            const int32_t n_outputs_new = lctx.n_outputs;
-
-            if (n_outputs_new) {
-                GGML_ASSERT( n_outputs_prev + n_outputs_new <= n_outputs);
-                GGML_ASSERT((n_outputs_prev + n_outputs_new)*n_vocab <= (int64_t) lctx.logits_size);
-                ggml_backend_tensor_get_async(backend_res, res, logits_out, 0, n_outputs_new*n_vocab*sizeof(float));
-            }
-        }
-
-        // extract embeddings
-        if (embd) {
-            ggml_backend_t backend_embd = ggml_backend_sched_get_tensor_backend(lctx.sched.get(), embd);
-            GGML_ASSERT(backend_embd != nullptr);
-
-            switch (cparams.pooling_type) {
-                case LLAMA_POOLING_TYPE_NONE:
-                    {
-                        // extract token embeddings
-                        GGML_ASSERT(lctx.embd != nullptr);
-                        float * embd_out = lctx.embd + n_outputs_prev*n_embd;
-                        const int32_t n_outputs_new = lctx.n_outputs;
-
-                        if (n_outputs_new) {
-                            GGML_ASSERT( n_outputs_prev + n_outputs_new <= n_outputs);
-                            GGML_ASSERT((n_outputs_prev + n_outputs_new)*n_embd <= (int64_t) lctx.embd_size);
-                            ggml_backend_tensor_get_async(backend_embd, embd, embd_out, 0, n_outputs_new*n_embd*sizeof(float));
-                        }
-                    } break;
-                case LLAMA_POOLING_TYPE_MEAN:
-                case LLAMA_POOLING_TYPE_CLS:
-                case LLAMA_POOLING_TYPE_LAST:
-                    {
-                        // extract sequence embeddings (cleared before processing each batch)
-                        auto & embd_seq_out = lctx.embd_seq;
-
-                        for (uint32_t s = 0; s < ubatch.n_seqs; ++s) {
-                            const llama_seq_id seq_id = ubatch.seq_id[s][0];
-                            if (embd_seq_out.find(seq_id) != embd_seq_out.end()) {
-                                continue;
-                            }
-                            embd_seq_out[seq_id].resize(n_embd);
-                            ggml_backend_tensor_get_async(backend_embd, embd, embd_seq_out[seq_id].data(), (n_embd*seq_id)*sizeof(float), n_embd*sizeof(float));
-                        }
-                    } break;
-                case LLAMA_POOLING_TYPE_RANK:
-                    {
-                        // extract the rerank score - a single float per sequence
-                        auto & embd_seq_out = lctx.embd_seq;
-
-                        for (uint32_t s = 0; s < ubatch.n_seqs; ++s) {
-                            const llama_seq_id seq_id = ubatch.seq_id[s][0];
-                            if (embd_seq_out.find(seq_id) != embd_seq_out.end()) {
-                                continue;
-                            }
-                            embd_seq_out[seq_id].resize(1);
-                            ggml_backend_tensor_get_async(backend_embd, embd, embd_seq_out[seq_id].data(), (seq_id)*sizeof(float), sizeof(float));
-                        }
-                    } break;
-                case LLAMA_POOLING_TYPE_UNSPECIFIED:
-                    {
-                        GGML_ABORT("unknown pooling type");
-                    }
-            }
-        }
-        n_outputs_prev += lctx.n_outputs;
-    }
-
-    // set output mappings
-    {
-        bool sorted_output = true;
-
-        GGML_ASSERT(lctx.sbatch.out_ids.size() == n_outputs);
-
-        for (size_t i = 0; i < n_outputs; ++i) {
-            size_t out_id = lctx.sbatch.out_ids[i];
-            lctx.output_ids[out_id] = i;
-            if (out_id != i) {
-                sorted_output = false;
-            }
-        }
-
-        if (sorted_output) {
-            lctx.sbatch.out_ids.clear();
-        }
-    }
-
-    // set to total number of outputs in the batch, for use in llama_get_logits_ith
-    lctx.n_outputs = n_outputs;
-
-    // wait for the computation to finish (automatically done when obtaining the model output)
-    //llama_synchronize(&lctx);
-
-    // decide if we need to defrag the kv cache
-    if (cparams.causal_attn && cparams.defrag_thold >= 0.0f) {
-        const float fragmentation = kv_self.n >= 128 ? 1.0f - float(kv_self.used)/float(kv_self.n) : 0.0f;
-
-        // queue defragmentation for next llama_kv_cache_update
-        if (fragmentation > cparams.defrag_thold) {
-            //LLAMA_LOG_INFO("fragmentation: %.2f\n", fragmentation);
-
-            llama_kv_cache_defrag(kv_self);
-        }
-    }
-
-    // Reset state for the next token before backend sync, to allow the CPU activities in the reset to
-    // overlap with device computation.
-    ggml_backend_sched_reset(lctx.sched.get());
-
-    return 0;
-}
-
-// encode a batch of tokens by evaluating the encoder part of the transformer
-//
-//   - lctx:      llama context
-//   - batch:     batch to evaluate
-//
-// return 0 on success
-// return positive int on warning
-// return negative int on error
-//
-static int llama_encode_internal(
-         llama_context & lctx,
-           llama_batch   inp_batch) {
-
-    lctx.is_encoding = true;
-
-    if (inp_batch.n_tokens == 0) {
-        LLAMA_LOG_ERROR("%s: n_tokens == 0\n", __func__);
-        return -1;
-    }
-
-    // temporary allocate memory for the input batch if needed
-    llama_batch_allocr batch_allocr(lctx, inp_batch);
-    const llama_batch & batch = batch_allocr.batch;
-    const uint32_t n_tokens = batch.n_tokens;
-
-    const auto & model   = lctx.model;
-    const auto & hparams = model.hparams;
-    const auto & cparams = lctx.cparams;
-
-    GGML_ASSERT((!batch.token && batch.embd) || (batch.token && !batch.embd)); // NOLINT
-
-    if (batch.token) {
-        for (uint32_t i = 0; i < n_tokens; ++i) {
-            if (batch.token[i] < 0 || (uint32_t)batch.token[i] >= model.vocab.n_vocab) {
-                LLAMA_LOG_ERROR("%s: invalid token[%d] = %d\n", __func__, i, batch.token[i]);
-                return -1;
-            }
-        }
-    }
-
-    // micro-batching is not possible for non-causal encoding, so we process the batch in a single shot
-    GGML_ASSERT(cparams.n_ubatch >= n_tokens && "encoder requires n_ubatch >= n_tokens");
-
-    if (lctx.t_compute_start_us == 0) {
-        lctx.t_compute_start_us = ggml_time_us();
-    }
-
-    lctx.n_queued_tokens += n_tokens;
-
-    const int64_t n_embd = hparams.n_embd;
-
-    lctx.sbatch.from_batch(batch, n_embd, /* simple_split */ true, /* logits_all */ true);
-
-    const llama_ubatch ubatch = lctx.sbatch.split_simple(n_tokens);
-
-    // reserve output buffer
-    if (llama_output_reserve(lctx, n_tokens) < n_tokens) {
-        LLAMA_LOG_ERROR("%s: could not reserve space for batch with %u outputs\n", __func__, n_tokens);
-        return -2;
-    };
-
-    for (uint32_t i = 0; i < n_tokens; ++i) {
-        lctx.output_ids[i] = i;
-    }
-
-    lctx.inp_embd_enc = NULL;
-    lctx.n_outputs = n_tokens;
-
-    int n_threads = n_tokens == 1 ? cparams.n_threads : cparams.n_threads_batch;
-    ggml_threadpool_t threadpool = n_tokens == 1 ? lctx.threadpool : lctx.threadpool_batch;
-
-    GGML_ASSERT(n_threads > 0);
-
-    ggml_backend_sched_reset(lctx.sched.get());
-    ggml_backend_sched_set_eval_callback(lctx.sched.get(), lctx.cparams.cb_eval, lctx.cparams.cb_eval_user_data);
-
-    ggml_cgraph * gf = llama_build_graph(lctx, ubatch, false);
-
-    // the output embeddings after the final encoder normalization
-    struct ggml_tensor * embd = nullptr;
-
-    // there are two cases here
-    if (llama_model_has_decoder(&lctx.model)) {
-        // first case is an encoder-decoder T5 model where embeddings are passed to decoder
-        embd = ggml_graph_node(gf, -1);
-        GGML_ASSERT(strcmp(embd->name, "result_norm") == 0 && "missing result_output tensor");
-    } else {
-        // second case is an encoder-only T5 model
-        if (cparams.embeddings) {
-            // only output embeddings if required
-            embd = ggml_graph_node(gf, -1);
-            if (strcmp(embd->name, "result_embd_pooled") != 0) {
-                embd = ggml_graph_node(gf, -2);
-            }
-            GGML_ASSERT(strcmp(embd->name, "result_embd_pooled") == 0 && "missing embeddings tensor");
-        }
-    }
-
-    ggml_backend_sched_alloc_graph(lctx.sched.get(), gf);
-
-    llama_set_inputs(lctx, ubatch);
-
-    const auto compute_status = llama_graph_compute(lctx, gf, n_threads, threadpool);
-    switch (compute_status) {
-        case GGML_STATUS_SUCCESS:
-            break;
-        case GGML_STATUS_ABORTED:
-            return 2;
-        case GGML_STATUS_ALLOC_FAILED:
-            return -2;
-        case GGML_STATUS_FAILED:
-        default:
-            return -3;
-    }
-
-    // extract embeddings
-    if (embd) {
-        ggml_backend_t backend_embd = ggml_backend_sched_get_tensor_backend(lctx.sched.get(), embd);
-        GGML_ASSERT(backend_embd != nullptr);
-
-        if (llama_model_has_decoder(&lctx.model)) {
-            lctx.embd_enc.resize(n_tokens*n_embd);
-            float * embd_out = lctx.embd_enc.data();
-
-            ggml_backend_tensor_get_async(backend_embd, embd, embd_out, 0, n_tokens*n_embd*sizeof(float));
-            GGML_ASSERT(!ubatch.equal_seqs); // TODO: handle equal splits
-
-            // remember the sequence ids used during the encoding - needed for cross attention later
-            lctx.seq_ids_enc.resize(n_tokens);
-            for (uint32_t i = 0; i < n_tokens; i++) {
-                for (int s = 0; s < ubatch.n_seq_id[i]; s++) {
-                    llama_seq_id seq_id = ubatch.seq_id[i][s];
-                    lctx.seq_ids_enc[i].insert(seq_id);
-                }
-            }
-        } else {
-            GGML_ASSERT(lctx.embd != nullptr);
-
-            switch (cparams.pooling_type) {
-                case LLAMA_POOLING_TYPE_NONE:
-                    {
-                        // extract token embeddings
-                        GGML_ASSERT(lctx.embd != nullptr);
-                        float * embd_out = lctx.embd;
-
-                        GGML_ASSERT(n_tokens*n_embd <= (int64_t) lctx.embd_size);
-                        ggml_backend_tensor_get_async(backend_embd, embd, embd_out, 0, n_tokens*n_embd*sizeof(float));
-                    } break;
-                case LLAMA_POOLING_TYPE_MEAN:
-                case LLAMA_POOLING_TYPE_CLS:
-                case LLAMA_POOLING_TYPE_LAST:
-                    {
-                        // extract sequence embeddings
-                        auto & embd_seq_out = lctx.embd_seq;
-                        embd_seq_out.clear();
-
-                        GGML_ASSERT(!ubatch.equal_seqs); // TODO: handle equal splits
-
-                        for (uint32_t i = 0; i < n_tokens; i++) {
-                            const llama_seq_id seq_id = ubatch.seq_id[i][0];
-                            if (embd_seq_out.find(seq_id) != embd_seq_out.end()) {
-                                continue;
-                            }
-                            embd_seq_out[seq_id].resize(n_embd);
-                            ggml_backend_tensor_get_async(backend_embd, embd, embd_seq_out[seq_id].data(), (n_embd*seq_id)*sizeof(float), n_embd*sizeof(float));
-                        }
-                    } break;
-                case LLAMA_POOLING_TYPE_RANK:
-                    {
-                        // TODO: this likely should be the same logic as in llama_decoder_internal, but better to
-                        //       wait for an encoder model that requires this pooling type in order to test it
-                        //       https://github.com/ggerganov/llama.cpp/pull/9510
-                        GGML_ABORT("RANK pooling not implemented yet");
-                    }
-                case LLAMA_POOLING_TYPE_UNSPECIFIED:
-                    {
-                        GGML_ABORT("unknown pooling type");
-                    }
-            }
-        }
-    }
-
-    // Reset state for the next token before backend sync, to allow the CPU activities in the reset to
-    // overlap with device computation.
-    ggml_backend_sched_reset(lctx.sched.get());
-
-    return 0;
-}
-
-// find holes from the beginning of the KV cache and fill them by moving data from the end of the cache
-static void llama_kv_cache_defrag_internal(struct llama_context & lctx) {
-    auto & kv_self = lctx.kv_self;
-
-    const auto & hparams = lctx.model.hparams;
-
-    const uint32_t n_layer = hparams.n_layer;
-
-    const uint32_t n_kv   = llama_kv_cache_cell_max(kv_self);
-    const uint32_t n_used = kv_self.used;
-
-    assert(n_used <= n_kv);
-
-    //const int64_t t_start = ggml_time_us();
-
-    // number of cells moved
-    uint32_t n_moves = 0;
-
-    // each move requires 6*n_layer tensors (see build_defrag)
-    //   - source view, destination view, copy operation
-    //   - x2 for keys and values
-    //const uint32_t max_moves = llama_model_max_nodes(model)/(6*n_layer);
-    // TODO: tmp fix https://github.com/ggerganov/llama.cpp/issues/6685#issuecomment-2057579516
-    const uint32_t max_moves = (llama_model_max_nodes(lctx.model) - 2*n_layer)/(6*n_layer);
-
-    // determine which KV cells to move where
-    //
-    //  cell i moves to ids[i]
-    //
-    //  if ids[i] == i || ids[i] == n_kv, then cell i is not moved
-    //
-    std::vector ids(n_kv, n_kv);
-
-    for (uint32_t i0 = 0; i0 < n_used; ++i0) {
-        const auto & cell0 = kv_self.cells[i0];
-
-        if (!cell0.is_empty()) {
-            ids[i0] = i0;
-
-            continue;
-        }
-
-        // found a hole - fill it with data from the end of the cache
-
-        uint32_t nh = 1;
-
-        // determine the size of the hole
-        while (i0 + nh < n_used && kv_self.cells[i0 + nh].is_empty()) {
-            nh++;
-        }
-
-        uint32_t nf = 0;
-        uint32_t is = n_kv - 1;
-
-        // starting from the end, find nh non-empty cells
-        for (; is > i0; --is) {
-            const auto & cell1 = kv_self.cells[is];
-
-            if (cell1.is_empty() || ids[is] != n_kv) {
-                continue;
-            }
-
-            // non-empty cell which is not yet moved
-            nf++;
-
-            if (nf == nh) {
-                break;
-            }
-        }
-
-        // this can only happen if `n_used` is not accurate, which would be a bug
-        GGML_ASSERT(nf == nh && "KV defrag bug: nf != nh");
-
-        nf = 0;
-
-        uint32_t i1 = is;
-
-        // are we moving a continuous block of memory?
-        bool cont = false;
-
-        // should we stop searching for the next move?
-        bool stop = false;
-
-        // go back and move the nf cells to the hole
-        for (; i1 < n_kv; ++i1) {
-            auto & cell1 = kv_self.cells[i1];
-
-            if (cell1.is_empty() || ids[i1] != n_kv) {
-                if (n_moves == max_moves) {
-                    stop = true;
-                    break;
-                }
-
-                cont = false;
-                continue;
-            }
-
-            // this cell goes to (i0 + nf)
-            ids[i1] = i0 + nf;
-
-            // move the cell meta data
-            kv_self.cells[i0 + nf] = cell1;
-
-            // clear the old cell and move the head there
-            cell1 = llama_kv_cell();
-            kv_self.head = n_used;
-
-            if (!cont) {
-                n_moves++;
-                cont = true;
-            }
-
-            nf++;
-
-            if (nf == nh) {
-                break;
-            }
-        }
-
-        if (stop || n_moves == max_moves) {
-            break;
-        }
-
-        //LLAMA_LOG_INFO("(tmp log) KV defrag: move [%u, %u) to [%u, %u)\n", is, i1 + 1, i0, i0 + nh);
-
-        i0 += nh - 1;
-    }
-
-    if (n_moves == 0) {
-        return;
-    }
-
-    //LLAMA_LOG_INFO("(tmp log) KV defrag cell moves: %u\n", n_moves);
-
-    //LLAMA_LOG_INFO("expected gf nodes: %u\n", 6*n_moves*n_layer);
-
-#if 0
-    // CPU defrag
-    //
-    // TODO: optimizations are possible:
-    //       - multiple threads
-    //       - avoid copying to the host memory when already there
-    //
-    // likely not worth the effort, as we have ggml_graph based defrag
-    //
-
-    const uint32_t n_embd_k_gqa = hparams.n_embd_k_gqa();
-    const uint32_t n_embd_v_gqa = hparams.n_embd_v_gqa();
-
-    const uint32_t kv_size = kv_self.size;
-
-    std::vector buf_k;
-    std::vector buf_v;
-
-    for (uint32_t il = 0; il < n_layer; ++il) {
-        const size_t k_size_row = ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa);
-        const size_t k_size     = ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa*kv_size);
-
-        const size_t v_size_el = ggml_type_size(kv_self.v_l[il]->type);
-        const size_t v_size    = ggml_row_size (kv_self.v_l[il]->type, n_embd_v_gqa*kv_size);
-
-        buf_k.resize(k_size);
-        buf_v.resize(v_size);
-
-        ggml_backend_tensor_get(kv_self.k_l[il], buf_k.data(), 0, buf_k.size());
-        ggml_backend_tensor_get(kv_self.v_l[il], buf_v.data(), 0, buf_v.size());
-
-        // batch move [i, i+nm) to [id, id+nm)
-        // note: cells can move only to a lower index
-        for (uint32_t i = 0; i < n_kv; ++i) {
-            const uint32_t id = ids[i];
-
-            if (i == id || id == n_kv) {
-                continue;
-            }
-
-            uint32_t nm = 1;
-
-            while (i + nm < n_kv && ids[i + nm] == id + nm) {
-                nm++;
-            }
-
-            // move keys
-            {
-                const int64_t os =  i*k_size_row;
-                const int64_t od = id*k_size_row;
-
-                memcpy(buf_k.data() + od, buf_k.data() + os, nm*k_size_row);
-            }
-
-            // move values (note: they are transposed)
-            {
-                const int64_t os =  i;
-                const int64_t od = id;
-
-                for (uint32_t j = 0; j < n_embd_v_gqa; ++j) {
-                    memcpy(buf_v.data() + (od + j*kv_size)*v_size_el, buf_v.data() + (os + j*kv_size)*v_size_el, nm*v_size_el);
-                }
-            }
-
-            i += nm - 1;
-        }
-
-        ggml_backend_tensor_set(kv_self.k_l[il], buf_k.data(), 0, buf_k.size());
-        ggml_backend_tensor_set(kv_self.v_l[il], buf_v.data(), 0, buf_v.size());
-    }
-#else
-    // ggml_graph defrag
-
-    ggml_backend_sched_reset(lctx.sched.get());
-
-    ggml_cgraph * gf = llama_build_graph_defrag(lctx, ids);
-
-    llama_graph_compute(lctx, gf, lctx.cparams.n_threads, lctx.threadpool);
-#endif
-
-    //const int64_t t_end = ggml_time_us();
-
-    //LLAMA_LOG_INFO("(tmp log) KV defrag time: %.3f ms\n", (t_end - t_start)/1000.0);
-}
-
-static void llama_kv_cache_update_internal(struct llama_context & lctx) {
-    bool need_reserve = false;
-
-    if (lctx.kv_self.has_shift) {
-        if (!llama_kv_cache_can_shift(&lctx)) {
-            GGML_ABORT("The current context does not support K-shift");
-        }
-
-        // apply K-shift if needed
-        if (lctx.model.hparams.rope_type != LLAMA_ROPE_TYPE_NONE) {
-            ggml_backend_sched_reset(lctx.sched.get());
-
-            ggml_cgraph * gf = llama_build_graph_k_shift(lctx);
-
-            ggml_backend_sched_alloc_graph(lctx.sched.get(), gf);
-
-            llama_set_k_shift(lctx);
-
-            llama_graph_compute(lctx, gf, lctx.cparams.n_threads, lctx.threadpool);
-
-            need_reserve = true;
-        }
-
-        {
-            auto & kv_self = lctx.kv_self;
-
-            kv_self.has_shift = false;
-
-            for (uint32_t i = 0; i < kv_self.size; ++i) {
-                kv_self.cells[i].delta = 0;
-            }
-        }
-    }
-
-    // defragment the KV cache if needed
-    if (lctx.kv_self.do_defrag) {
-        llama_kv_cache_defrag_internal(lctx);
-
-        need_reserve = true;
-
-        lctx.kv_self.do_defrag = false;
-    }
-
-    // reserve a worst case graph again
-    if (need_reserve) {
-        // TODO: extract to a function
-        // build worst-case graph
-        uint32_t n_seqs = 1; // TODO: worst-case number of sequences
-        uint32_t n_tokens = std::min(lctx.cparams.n_ctx, lctx.cparams.n_ubatch);
-        llama_token token = llama_token_bos(&lctx.model); // not actually used by llama_build_graph, but required to choose between token and embedding inputs graph
-        llama_ubatch ubatch = { true, n_tokens, n_tokens / n_seqs, n_seqs, &token, nullptr, nullptr, nullptr, nullptr, nullptr};
-        ggml_cgraph * gf = llama_build_graph(lctx, ubatch, true);
-
-        // initialize scheduler with the worst-case graph
-        ggml_backend_sched_reset(lctx.sched.get());
-        if (!ggml_backend_sched_reserve(lctx.sched.get(), gf)) {
-            LLAMA_LOG_ERROR("%s: failed to allocate compute buffers\n", __func__);
-        }
-    }
-}
-
-//
-// quantization
-//
-
-struct quantize_state_internal {
-    const llama_model                 & model;
-    const llama_model_quantize_params * params;
-
-    int n_attention_wv    = 0;
-    int n_ffn_down        = 0;
-    int n_ffn_gate        = 0;
-    int n_ffn_up          = 0;
-    int i_attention_wv    = 0;
-    int i_ffn_down        = 0;
-    int i_ffn_gate        = 0;
-    int i_ffn_up          = 0;
-
-    int n_k_quantized     = 0;
-    int n_fallback        = 0;
-
-    bool has_imatrix      = false;
-
-    // used to figure out if a model shares tok_embd with the output weight
-    bool has_output       = false;
-
-    quantize_state_internal(const llama_model & model, const llama_model_quantize_params * params)
-        : model(model)
-        , params(params)
-        {}
-};
-
-static void llama_tensor_dequantize_internal(
-    struct ggml_tensor * tensor, std::vector> & output, std::vector & workers,
-    const size_t nelements, const int nthread
-) {
-    if (output.size() < nelements) {
-        output.resize(nelements);
-    }
-    float * f32_output = (float *) output.data();
-
-    const ggml_type_traits * qtype = ggml_get_type_traits(tensor->type);
-    if (ggml_is_quantized(tensor->type)) {
-        if (qtype->to_float == NULL) {
-            throw std::runtime_error(format("type %s unsupported for integer quantization: no dequantization available", ggml_type_name(tensor->type)));
-        }
-    } else if (tensor->type != GGML_TYPE_F16 &&
-               tensor->type != GGML_TYPE_BF16) {
-        throw std::runtime_error(format("cannot dequantize/convert tensor type %s", ggml_type_name(tensor->type)));
-    }
-
-    if (nthread < 2) {
-        if (tensor->type == GGML_TYPE_F16) {
-            ggml_fp16_to_fp32_row((ggml_fp16_t *)tensor->data, f32_output, nelements);
-        } else if (tensor->type == GGML_TYPE_BF16) {
-            ggml_bf16_to_fp32_row((ggml_bf16_t *)tensor->data, f32_output, nelements);
-        } else if (ggml_is_quantized(tensor->type)) {
-            qtype->to_float(tensor->data, f32_output, nelements);
-        } else {
-            GGML_ABORT("fatal error"); // unreachable
-        }
-        return;
-    }
-
-    size_t block_size;
-    if (tensor->type == GGML_TYPE_F16 ||
-        tensor->type == GGML_TYPE_BF16) {
-        block_size = 1;
-    } else {
-        block_size = (size_t)ggml_blck_size(tensor->type);
-    }
-
-    size_t block_size_bytes = ggml_type_size(tensor->type);
-
-    GGML_ASSERT(nelements % block_size == 0);
-    size_t nblocks = nelements / block_size;
-    size_t blocks_per_thread = nblocks / nthread;
-    size_t spare_blocks = nblocks - (blocks_per_thread * nthread); // if blocks aren't divisible by thread count
-
-    size_t in_buff_offs = 0;
-    size_t out_buff_offs = 0;
-
-    for (int tnum = 0; tnum < nthread; tnum++) {
-        size_t thr_blocks = blocks_per_thread + (tnum == nthread - 1 ? spare_blocks : 0); // num blocks for this thread
-        size_t thr_elems = thr_blocks * block_size; // number of elements for this thread
-        size_t thr_block_bytes = thr_blocks * block_size_bytes; // number of input bytes for this thread
-
-        auto compute = [qtype] (ggml_type typ, uint8_t * inbuf, float * outbuf, int nels) {
-            if (typ == GGML_TYPE_F16) {
-                ggml_fp16_to_fp32_row((ggml_fp16_t *)inbuf, outbuf, nels);
-            } else if (typ == GGML_TYPE_BF16) {
-                ggml_bf16_to_fp32_row((ggml_bf16_t *)inbuf, outbuf, nels);
-            } else {
-                qtype->to_float(inbuf, outbuf, nels);
-            }
-        };
-        workers.emplace_back(compute, tensor->type, (uint8_t *) tensor->data + in_buff_offs, f32_output + out_buff_offs, thr_elems);
-        in_buff_offs += thr_block_bytes;
-        out_buff_offs += thr_elems;
-    }
-    for (auto & w : workers) { w.join(); }
-    workers.clear();
-}
-
-static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type new_type, const ggml_tensor * tensor, llama_ftype ftype) {
-    const std::string name = ggml_get_name(tensor);
-
-    // TODO: avoid hardcoded tensor names - use the TN_* constants
-    const llm_arch arch = qs.model.arch;
-    const auto       tn = LLM_TN(arch);
-
-    auto use_more_bits = [](int i_layer, int n_layers) -> bool {
-        return i_layer < n_layers/8 || i_layer >= 7*n_layers/8 || (i_layer - n_layers/8)%3 == 2;
-    };
-    const int n_expert = std::max(1, (int)qs.model.hparams.n_expert);
-    auto layer_info = [n_expert] (int i_layer, int n_layer, const char * name) {
-        if (n_expert > 1) {
-            // Believe it or not, "experts" in the FFN of Mixtral-8x7B are not consecutive, but occasionally randomly
-            // sprinkled in the model. Hence, simply dividing i_ffn_down by n_expert does not work
-            // for getting the current layer as I initially thought, and we need to resort to parsing the
-            // tensor name.
-            if (sscanf(name, "blk.%d.", &i_layer) != 1) {
-                throw std::runtime_error(format("Failed to determine layer for tensor %s", name));
-            }
-            if (i_layer < 0 || i_layer >= n_layer) {
-                throw std::runtime_error(format("Bad layer %d for tensor %s. Must be in [0, %d)", i_layer, name, n_layer));
-            }
-        }
-        return std::make_pair(i_layer, n_layer);
-    };
-
-    // for arches that share the same tensor between the token embeddings and the output, we quantize the token embeddings
-    // with the quantization of the output tensor
-    if (name == tn(LLM_TENSOR_OUTPUT, "weight") || (!qs.has_output && name == tn(LLM_TENSOR_TOKEN_EMBD, "weight"))) {
-        if (qs.params->output_tensor_type < GGML_TYPE_COUNT) {
-            new_type = qs.params->output_tensor_type;
-        } else {
-            int nx = tensor->ne[0];
-            if (arch == LLM_ARCH_FALCON || nx % QK_K != 0) {
-                new_type = GGML_TYPE_Q8_0;
-            }
-            else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_XXS || ftype == LLAMA_FTYPE_MOSTLY_IQ2_XS || ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS ||
-                     ftype == LLAMA_FTYPE_MOSTLY_IQ1_S   || ftype == LLAMA_FTYPE_MOSTLY_IQ2_S  || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M   ||
-                     ftype == LLAMA_FTYPE_MOSTLY_IQ1_M) {
-                new_type = GGML_TYPE_Q5_K;
-            }
-            else if (new_type != GGML_TYPE_Q8_0) {
-                new_type = GGML_TYPE_Q6_K;
-            }
-        }
-    } else if (name == "token_embd.weight") {
-        if (qs.params->token_embedding_type < GGML_TYPE_COUNT) {
-            new_type = qs.params->token_embedding_type;
-        } else {
-            if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_XXS || ftype == LLAMA_FTYPE_MOSTLY_IQ2_XS ||
-                ftype == LLAMA_FTYPE_MOSTLY_IQ1_S   || ftype == LLAMA_FTYPE_MOSTLY_IQ1_M) {
-                new_type = GGML_TYPE_Q2_K;
-            }
-            else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M) {
-                new_type = GGML_TYPE_IQ3_S;
-            }
-            else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS) {
-                new_type = GGML_TYPE_IQ3_S;
-            }
-            else if (ftype == LLAMA_FTYPE_MOSTLY_TQ1_0 || ftype == LLAMA_FTYPE_MOSTLY_TQ2_0) {
-                new_type = GGML_TYPE_Q4_K;
-            }
-        }
-    } else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_XXS || ftype == LLAMA_FTYPE_MOSTLY_IQ2_XS || ftype == LLAMA_FTYPE_MOSTLY_IQ1_S ||
-               ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M    || ftype == LLAMA_FTYPE_MOSTLY_IQ1_M) {
-        if (name.find("attn_v.weight") != std::string::npos) {
-            if (qs.model.hparams.n_gqa() >= 4 || qs.model.hparams.n_expert >= 4) new_type = GGML_TYPE_Q4_K;
-            else new_type = ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M ? GGML_TYPE_IQ3_S : GGML_TYPE_Q2_K;
-            ++qs.i_attention_wv;
-        }
-        else if (qs.model.hparams.n_expert == 8 && name.find("attn_k.weight") != std::string::npos) {
-            new_type = GGML_TYPE_Q4_K;
-        }
-        else if (name.find("ffn_down") != std::string::npos) {
-            if (qs.i_ffn_down < qs.n_ffn_down/8) {
-                new_type = ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M ? GGML_TYPE_IQ3_S : GGML_TYPE_Q2_K;
-            }
-            ++qs.i_ffn_down;
-        }
-        else if (name.find("attn_output.weight") != std::string::npos) {
-            if (qs.model.hparams.n_expert == 8) {
-                new_type = GGML_TYPE_Q5_K;
-            } else {
-                if (ftype == LLAMA_FTYPE_MOSTLY_IQ1_S || ftype == LLAMA_FTYPE_MOSTLY_IQ1_M) new_type = GGML_TYPE_IQ2_XXS;
-                else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M) new_type = GGML_TYPE_IQ3_S;
-            }
-        }
-    } else if (name.find("attn_v.weight") != std::string::npos) {
-        if      (ftype == LLAMA_FTYPE_MOSTLY_Q2_K) {
-            new_type = qs.model.hparams.n_gqa() >= 4 ? GGML_TYPE_Q4_K : GGML_TYPE_Q3_K;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S && qs.model.hparams.n_gqa() >= 4) {
-            new_type = GGML_TYPE_Q4_K;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS) {
-            new_type = qs.model.hparams.n_gqa() >= 4 ? GGML_TYPE_Q4_K : !qs.has_imatrix ? GGML_TYPE_IQ3_S : GGML_TYPE_IQ3_XXS;
-        }
-        else if ((ftype == LLAMA_FTYPE_MOSTLY_IQ3_XS || ftype == LLAMA_FTYPE_MOSTLY_IQ3_S) && qs.model.hparams.n_gqa() >= 4) {
-            new_type = GGML_TYPE_Q4_K;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_M) {
-            new_type = GGML_TYPE_Q4_K;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_M) {
-            new_type = qs.i_attention_wv < 2 ? GGML_TYPE_Q5_K : GGML_TYPE_Q4_K;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_L) new_type = GGML_TYPE_Q5_K;
-        else if ((ftype == LLAMA_FTYPE_MOSTLY_IQ4_NL || ftype == LLAMA_FTYPE_MOSTLY_IQ4_XS) && qs.model.hparams.n_gqa() >= 4) {
-            new_type = GGML_TYPE_Q5_K;
-        }
-        else if ((ftype == LLAMA_FTYPE_MOSTLY_Q4_K_M || ftype == LLAMA_FTYPE_MOSTLY_Q5_K_M) &&
-                use_more_bits(qs.i_attention_wv, qs.n_attention_wv)) new_type = GGML_TYPE_Q6_K;
-        else if (ftype == LLAMA_FTYPE_MOSTLY_Q4_K_S && qs.i_attention_wv < 4) new_type = GGML_TYPE_Q5_K;
-        if (qs.model.type == MODEL_70B) {
-            // In the 70B model we have 8 heads sharing the same attn_v weights. As a result, the attn_v.weight tensor is
-            // 8x smaller compared to attn_q.weight. Hence, we can get a nice boost in quantization accuracy with
-            // nearly negligible increase in model size by quantizing this tensor with more bits:
-            if (new_type == GGML_TYPE_Q3_K || new_type == GGML_TYPE_Q4_K) new_type = GGML_TYPE_Q5_K;
-        }
-        if (qs.model.hparams.n_expert == 8) {
-            // for the 8-expert model, bumping this to Q8_0 trades just ~128MB
-            // TODO: explore better strategies
-            new_type = GGML_TYPE_Q8_0;
-        }
-        ++qs.i_attention_wv;
-    } else if (name.find("attn_k.weight") != std::string::npos) {
-        if (qs.model.hparams.n_expert == 8) {
-            // for the 8-expert model, bumping this to Q8_0 trades just ~128MB
-            // TODO: explore better strategies
-            new_type = GGML_TYPE_Q8_0;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XS) {
-            new_type = GGML_TYPE_IQ3_XXS;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS) {
-            new_type = GGML_TYPE_IQ2_S;
-        }
-    } else if (name.find("attn_q.weight") != std::string::npos) {
-        if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XS) {
-            new_type = GGML_TYPE_IQ3_XXS;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS) {
-            new_type = GGML_TYPE_IQ2_S;
-        }
-    } else if (name.find("ffn_down") != std::string::npos) {
-        auto info = layer_info(qs.i_ffn_down, qs.n_ffn_down, name.c_str());
-        int i_layer = info.first, n_layer = info.second;
-        if      (ftype == LLAMA_FTYPE_MOSTLY_Q2_K) new_type = GGML_TYPE_Q3_K;
-        else if (ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S) {
-            if (i_layer < n_layer/8) new_type = GGML_TYPE_Q4_K;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS && !qs.has_imatrix) {
-            new_type = i_layer < n_layer/8 ? GGML_TYPE_Q4_K : GGML_TYPE_Q3_K;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_M) {
-            new_type = i_layer < n_layer/16 ? GGML_TYPE_Q5_K
-                     : arch != LLM_ARCH_FALCON || use_more_bits(i_layer, n_layer) ? GGML_TYPE_Q4_K
-                     : GGML_TYPE_Q3_K;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_M && (i_layer < n_layer/8 ||
-                    (qs.model.hparams.n_expert == 8 && use_more_bits(i_layer, n_layer)))) {
-            new_type = GGML_TYPE_Q4_K;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_L) {
-            new_type = arch == LLM_ARCH_FALCON ? GGML_TYPE_Q4_K : GGML_TYPE_Q5_K;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_Q4_K_M) {
-            if (arch == LLM_ARCH_FALCON) {
-                new_type = i_layer < n_layer/16 ? GGML_TYPE_Q6_K :
-                           use_more_bits(i_layer, n_layer) ? GGML_TYPE_Q5_K : GGML_TYPE_Q4_K;
-            } else {
-                if (use_more_bits(i_layer, n_layer)) new_type = GGML_TYPE_Q6_K;
-            }
-        }
-        else if (i_layer < n_layer/8 && (ftype == LLAMA_FTYPE_MOSTLY_IQ4_NL || ftype == LLAMA_FTYPE_MOSTLY_IQ4_XS) && !qs.has_imatrix) {
-            new_type = GGML_TYPE_Q5_K;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_Q5_K_M && use_more_bits(i_layer, n_layer)) new_type = GGML_TYPE_Q6_K;
-        else if (ftype == LLAMA_FTYPE_MOSTLY_Q4_K_S && arch != LLM_ARCH_FALCON && i_layer < n_layer/8) {
-            new_type = GGML_TYPE_Q5_K;
-        }
-        else if ((ftype == LLAMA_FTYPE_MOSTLY_Q4_0 || ftype == LLAMA_FTYPE_MOSTLY_Q5_0)
-                && qs.has_imatrix && i_layer < n_layer/8) {
-            // Guard against craziness in the first few ffn_down layers that can happen even with imatrix for Q4_0/Q5_0.
-            // We only do it when an imatrix is provided because a) we want to make sure that one can always get the
-            // same quantization as before imatrix stuff, and b) Q4_1/Q5_1 do go crazy on ffn_down without an imatrix.
-            new_type = ftype == LLAMA_FTYPE_MOSTLY_Q4_0 ? GGML_TYPE_Q4_1 : GGML_TYPE_Q5_1;
-        }
-        ++qs.i_ffn_down;
-    } else if (name.find("attn_output.weight") != std::string::npos) {
-        if (arch != LLM_ARCH_FALCON) {
-            if (qs.model.hparams.n_expert == 8) {
-                if (ftype == LLAMA_FTYPE_MOSTLY_Q2_K   || ftype == LLAMA_FTYPE_MOSTLY_IQ3_XS || ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS ||
-                    ftype == LLAMA_FTYPE_MOSTLY_Q3_K_S || ftype == LLAMA_FTYPE_MOSTLY_Q3_K_M  || ftype == LLAMA_FTYPE_MOSTLY_IQ4_NL  ||
-                    ftype == LLAMA_FTYPE_MOSTLY_Q4_K_S || ftype == LLAMA_FTYPE_MOSTLY_Q4_K_M  || ftype == LLAMA_FTYPE_MOSTLY_IQ3_S  ||
-                    ftype == LLAMA_FTYPE_MOSTLY_IQ3_M  || ftype == LLAMA_FTYPE_MOSTLY_IQ4_XS) {
-                    new_type = GGML_TYPE_Q5_K;
-                }
-            } else {
-                if      (ftype == LLAMA_FTYPE_MOSTLY_Q2_K   ) new_type = GGML_TYPE_Q3_K;
-                else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS) new_type = GGML_TYPE_IQ3_S;
-                else if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_M ) new_type = GGML_TYPE_Q4_K;
-                else if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_L ) new_type = GGML_TYPE_Q5_K;
-                else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_M  ) new_type = GGML_TYPE_Q4_K;
-            }
-        } else {
-            if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_L) new_type = GGML_TYPE_Q4_K;
-        }
-    }
-    else if (name.find("attn_qkv.weight") != std::string::npos) {
-        if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_M || ftype == LLAMA_FTYPE_MOSTLY_Q3_K_L || ftype == LLAMA_FTYPE_MOSTLY_IQ3_M) {
-            new_type = GGML_TYPE_Q4_K;
-        }
-        else if (ftype == LLAMA_FTYPE_MOSTLY_Q4_K_M) new_type = GGML_TYPE_Q5_K;
-        else if (ftype == LLAMA_FTYPE_MOSTLY_Q5_K_M) new_type = GGML_TYPE_Q6_K;
-    }
-    else if (name.find("ffn_gate") != std::string::npos) {
-        auto info = layer_info(qs.i_ffn_gate, qs.n_ffn_gate, name.c_str());
-        int i_layer = info.first, n_layer = info.second;
-        if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XS && (i_layer >= n_layer/8 && i_layer < 7*n_layer/8)) {
-            new_type = GGML_TYPE_IQ3_XXS;
-        }
-        ++qs.i_ffn_gate;
-    }
-    else if (name.find("ffn_up") != std::string::npos) {
-        auto info = layer_info(qs.i_ffn_up, qs.n_ffn_up, name.c_str());
-        int i_layer = info.first, n_layer = info.second;
-        if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XS && (i_layer >= n_layer/8 && i_layer < 7*n_layer/8)) {
-            new_type = GGML_TYPE_IQ3_XXS;
-        }
-        ++qs.i_ffn_up;
-    }
-
-    //    if (ftype == LLAMA_FTYPE_MOSTLY_Q2_K) new_type = GGML_TYPE_Q3_K;
-    //}
-    // IK: let's remove this, else Q2_K is almost the same as Q3_K_S
-    //else if (name.find("ffn_gate") != std::string::npos || name.find("ffn_up") != std::string::npos) {
-    //    if (ftype == LLAMA_FTYPE_MOSTLY_Q2_K) new_type = GGML_TYPE_Q3_K;
-    //}
-    // This can be used to reduce the size of the Q5_K_S model.
-    // The associated PPL increase is fully in line with the size reduction
-    //else {
-    //    if (ftype == LLAMA_FTYPE_MOSTLY_Q5_K_S) new_type = GGML_TYPE_Q4_K;
-    //}
-    bool convert_incompatible_tensor = false;
-    if (new_type == GGML_TYPE_Q2_K    || new_type == GGML_TYPE_Q3_K    || new_type == GGML_TYPE_Q4_K   ||
-        new_type == GGML_TYPE_Q5_K    || new_type == GGML_TYPE_Q6_K    || new_type == GGML_TYPE_IQ4_XS ||
-        new_type == GGML_TYPE_IQ2_XS  || new_type == GGML_TYPE_IQ2_XXS || new_type == GGML_TYPE_IQ2_S  ||
-        new_type == GGML_TYPE_IQ3_XXS || new_type == GGML_TYPE_IQ1_S   || new_type == GGML_TYPE_IQ3_S  ||
-        new_type == GGML_TYPE_IQ1_M) {
-        int nx = tensor->ne[0];
-        int ny = tensor->ne[1];
-        if (nx % QK_K != 0) {
-            LLAMA_LOG_WARN("\n\n%s : tensor cols %d x %d are not divisible by %d, required for %s", __func__, nx, ny, QK_K, ggml_type_name(new_type));
-            convert_incompatible_tensor = true;
-        } else {
-            ++qs.n_k_quantized;
-        }
-    }
-    if (convert_incompatible_tensor) {
-        switch (new_type) {
-            case GGML_TYPE_TQ1_0:
-            case GGML_TYPE_TQ2_0:  new_type = GGML_TYPE_Q4_0; break;  // TODO: use a symmetric type instead
-            case GGML_TYPE_IQ2_XXS:
-            case GGML_TYPE_IQ2_XS:
-            case GGML_TYPE_IQ2_S:
-            case GGML_TYPE_IQ3_XXS:
-            case GGML_TYPE_IQ3_S:
-            case GGML_TYPE_IQ1_S:
-            case GGML_TYPE_IQ1_M:
-            case GGML_TYPE_Q2_K:
-            case GGML_TYPE_Q3_K:
-            case GGML_TYPE_IQ4_XS: new_type = GGML_TYPE_IQ4_NL; break;
-            case GGML_TYPE_Q4_K:   new_type = GGML_TYPE_Q5_0;   break;
-            case GGML_TYPE_Q5_K:   new_type = GGML_TYPE_Q5_1;   break;
-            case GGML_TYPE_Q6_K:   new_type = GGML_TYPE_Q8_0;   break;
-            default: throw std::runtime_error("\nUnsupported tensor size encountered\n");
-        }
-        if (tensor->ne[0] % ggml_blck_size(new_type) != 0) {
-            new_type = GGML_TYPE_F16;
-        }
-        LLAMA_LOG_WARN(" - using fallback quantization %s\n", ggml_type_name(new_type));
-        ++qs.n_fallback;
-    }
-
-    return new_type;
-}
-
-static size_t llama_tensor_quantize_internal(enum ggml_type new_type, const float * f32_data, void * new_data, const int64_t chunk_size, int64_t nrows, int64_t n_per_row, const float * imatrix, std::vector & workers, const int nthread) {
-    if (nthread < 2) {
-        // single-thread
-        size_t new_size = ggml_quantize_chunk(new_type, f32_data, new_data, 0, nrows, n_per_row, imatrix);
-        if (!ggml_validate_row_data(new_type, new_data, new_size)) {
-            throw std::runtime_error("quantized data validation failed");
-        }
-        return new_size;
-    }
-
-    std::mutex mutex;
-    int64_t counter = 0;
-    size_t new_size = 0;
-    bool valid = true;
-    auto compute = [&mutex, &counter, &new_size, &valid, new_type, f32_data, new_data, chunk_size,
-            nrows, n_per_row, imatrix]() {
-        const int64_t nrows_per_chunk = chunk_size / n_per_row;
-        size_t local_size = 0;
-        while (true) {
-            std::unique_lock lock(mutex);
-            int64_t first_row = counter; counter += nrows_per_chunk;
-            if (first_row >= nrows) {
-                if (local_size > 0) {
-                    new_size += local_size;
-                }
-                break;
-            }
-            lock.unlock();
-            const int64_t this_nrow = std::min(nrows - first_row, nrows_per_chunk);
-            size_t this_size = ggml_quantize_chunk(new_type, f32_data, new_data, first_row * n_per_row, this_nrow, n_per_row, imatrix);
-            local_size += this_size;
-
-            // validate the quantized data
-            const size_t row_size  = ggml_row_size(new_type, n_per_row);
-            void * this_data = (char *) new_data + first_row * row_size;
-            if (!ggml_validate_row_data(new_type, this_data, this_size)) {
-                std::unique_lock lock(mutex);
-                valid = false;
-                break;
-            }
-        }
-    };
-    for (int it = 0; it < nthread - 1; ++it) {
-        workers.emplace_back(compute);
-    }
-    compute();
-    for (auto & w : workers) { w.join(); }
-    workers.clear();
-    if (!valid) {
-        throw std::runtime_error("quantized data validation failed");
-    }
-    return new_size;
-}
-
-static void llama_model_quantize_internal(const std::string & fname_inp, const std::string & fname_out, const llama_model_quantize_params * params) {
-    ggml_type default_type;
-    llama_ftype ftype = params->ftype;
-
-    switch (params->ftype) {
-        case LLAMA_FTYPE_MOSTLY_Q4_0: default_type = GGML_TYPE_Q4_0; break;
-        case LLAMA_FTYPE_MOSTLY_Q4_1: default_type = GGML_TYPE_Q4_1; break;
-        case LLAMA_FTYPE_MOSTLY_Q5_0: default_type = GGML_TYPE_Q5_0; break;
-        case LLAMA_FTYPE_MOSTLY_Q5_1: default_type = GGML_TYPE_Q5_1; break;
-        case LLAMA_FTYPE_MOSTLY_Q8_0: default_type = GGML_TYPE_Q8_0; break;
-        case LLAMA_FTYPE_MOSTLY_F16:  default_type = GGML_TYPE_F16;  break;
-        case LLAMA_FTYPE_MOSTLY_BF16: default_type = GGML_TYPE_BF16; break;
-        case LLAMA_FTYPE_ALL_F32:     default_type = GGML_TYPE_F32;  break;
-
-        // K-quants
-        case LLAMA_FTYPE_MOSTLY_Q2_K_S:
-        case LLAMA_FTYPE_MOSTLY_Q2_K:    default_type = GGML_TYPE_Q2_K;    break;
-        case LLAMA_FTYPE_MOSTLY_IQ3_XS:  default_type = GGML_TYPE_IQ3_S;   break;
-        case LLAMA_FTYPE_MOSTLY_Q3_K_S:
-        case LLAMA_FTYPE_MOSTLY_Q3_K_M:
-        case LLAMA_FTYPE_MOSTLY_Q3_K_L:  default_type = GGML_TYPE_Q3_K;    break;
-        case LLAMA_FTYPE_MOSTLY_Q4_K_S:
-        case LLAMA_FTYPE_MOSTLY_Q4_K_M:  default_type = GGML_TYPE_Q4_K;    break;
-        case LLAMA_FTYPE_MOSTLY_Q5_K_S:
-        case LLAMA_FTYPE_MOSTLY_Q5_K_M:  default_type = GGML_TYPE_Q5_K;    break;
-        case LLAMA_FTYPE_MOSTLY_Q6_K:    default_type = GGML_TYPE_Q6_K;    break;
-        case LLAMA_FTYPE_MOSTLY_TQ1_0:   default_type = GGML_TYPE_TQ1_0;   break;
-        case LLAMA_FTYPE_MOSTLY_TQ2_0:   default_type = GGML_TYPE_TQ2_0;   break;
-        case LLAMA_FTYPE_MOSTLY_IQ2_XXS: default_type = GGML_TYPE_IQ2_XXS; break;
-        case LLAMA_FTYPE_MOSTLY_IQ2_XS:  default_type = GGML_TYPE_IQ2_XS;  break;
-        case LLAMA_FTYPE_MOSTLY_IQ2_S:   default_type = GGML_TYPE_IQ2_XS;  break;
-        case LLAMA_FTYPE_MOSTLY_IQ2_M:   default_type = GGML_TYPE_IQ2_S;   break;
-        case LLAMA_FTYPE_MOSTLY_IQ3_XXS: default_type = GGML_TYPE_IQ3_XXS; break;
-        case LLAMA_FTYPE_MOSTLY_IQ1_S:   default_type = GGML_TYPE_IQ1_S;   break;
-        case LLAMA_FTYPE_MOSTLY_IQ1_M:   default_type = GGML_TYPE_IQ1_M;   break;
-        case LLAMA_FTYPE_MOSTLY_IQ4_NL:  default_type = GGML_TYPE_IQ4_NL;  break;
-        case LLAMA_FTYPE_MOSTLY_IQ4_XS:  default_type = GGML_TYPE_IQ4_XS;  break;
-        case LLAMA_FTYPE_MOSTLY_IQ3_S:   default_type = GGML_TYPE_IQ3_S;   break;
-        case LLAMA_FTYPE_MOSTLY_IQ3_M:   default_type = GGML_TYPE_IQ3_S;   break;
-
-        default: throw std::runtime_error(format("invalid output file type %d\n", ftype));
-    }
-
-    int nthread = params->nthread;
-
-    if (nthread <= 0) {
-        nthread = std::thread::hardware_concurrency();
-    }
-
-    // mmap consistently increases speed Linux, and also increases speed on Windows with
-    // hot cache. It may cause a slowdown on macOS, possibly related to free memory.
-#if defined(__linux__) || defined(_WIN32)
-    constexpr bool use_mmap = true;
-#else
-    constexpr bool use_mmap = false;
-#endif
-
-    llama_model_kv_override * kv_overrides = nullptr;
-    if (params->kv_overrides) {
-        auto v = (std::vector*)params->kv_overrides;
-        kv_overrides = v->data();
-    }
-    llama_model_loader ml(fname_inp, use_mmap, /*check_tensors*/ true, kv_overrides);
-    ml.init_mappings(false); // no prefetching
-
-    llama_model model;
-    llm_load_arch(ml, model);
-    llm_load_hparams(ml, model);
-    llm_load_stats(ml, model);
-
-    struct quantize_state_internal qs(model, params);
-
-    if (params->only_copy) {
-        ftype = model.ftype;
-    }
-    const std::unordered_map> * imatrix_data = nullptr;
-    if (params->imatrix) {
-        imatrix_data = static_cast>*>(params->imatrix);
-        if (imatrix_data) {
-            LLAMA_LOG_INFO("================================ Have weights data with %d entries\n",int(imatrix_data->size()));
-            qs.has_imatrix = true;
-            // check imatrix for nans or infs
-            for (const auto & kv : *imatrix_data) {
-                for (float f : kv.second) {
-                    if (!std::isfinite(f)) {
-                        throw std::runtime_error(format("imatrix contains non-finite value %f\n", f));
-                    }
-                }
-            }
-        }
-    }
-
-    const size_t align = GGUF_DEFAULT_ALIGNMENT;
-    gguf_context_ptr ctx_out { gguf_init_empty() };
-
-    // copy the KV pairs from the input file
-    gguf_set_kv     (ctx_out.get(), ml.meta.get());
-    gguf_set_val_u32(ctx_out.get(), "general.quantization_version", GGML_QNT_VERSION); // TODO: use LLM_KV
-    gguf_set_val_u32(ctx_out.get(), "general.file_type", ftype); // TODO: use LLM_KV
-
-    // Remove split metadata
-    gguf_remove_key(ctx_out.get(), ml.llm_kv(LLM_KV_SPLIT_NO).c_str());
-    gguf_remove_key(ctx_out.get(), ml.llm_kv(LLM_KV_SPLIT_COUNT).c_str());
-    gguf_remove_key(ctx_out.get(), ml.llm_kv(LLM_KV_SPLIT_TENSORS_COUNT).c_str());
-
-    if (params->kv_overrides) {
-        const std::vector & overrides = *(const std::vector *)params->kv_overrides;
-        for (const auto & o : overrides) {
-            if (o.key[0] == 0) break;
-            if (o.tag == LLAMA_KV_OVERRIDE_TYPE_FLOAT) {
-                gguf_set_val_f32(ctx_out.get(), o.key, o.val_f64);
-            } else if (o.tag == LLAMA_KV_OVERRIDE_TYPE_INT) {
-                gguf_set_val_i32(ctx_out.get(), o.key, o.val_i64);
-            } else if (o.tag == LLAMA_KV_OVERRIDE_TYPE_BOOL) {
-                gguf_set_val_bool(ctx_out.get(), o.key, o.val_bool);
-            } else if (o.tag == LLAMA_KV_OVERRIDE_TYPE_STR) {
-                gguf_set_val_str(ctx_out.get(), o.key, o.val_str);
-            } else {
-                LLAMA_LOG_WARN("%s: unknown KV override type for key %s\n", __func__, o.key);
-            }
-        }
-    }
-
-    // make a list of weights
-    std::vector tensors;
-    tensors.reserve(ml.weights_map.size());
-    for (const auto & it : ml.weights_map) {
-        tensors.push_back(&it.second);
-    }
-
-    // keep_split requires that the weights are sorted by split index
-    if (params->keep_split) {
-        std::sort(tensors.begin(), tensors.end(), [](const llama_model_loader::llama_tensor_weight * a, const llama_model_loader::llama_tensor_weight * b) {
-            if (a->idx == b->idx) {
-                return a->offs < b->offs;
-            }
-            return a->idx < b->idx;
-        });
-    }
-
-    for (const auto * it : tensors) {
-        const struct ggml_tensor * tensor = it->tensor;
-
-        const std::string name = ggml_get_name(tensor);
-
-        // TODO: avoid hardcoded tensor names - use the TN_* constants
-        if (name.find("attn_v.weight")   != std::string::npos ||
-            name.find("attn_qkv.weight") != std::string::npos ||
-            name.find("attn_kv_b.weight")!= std::string::npos) {
-            ++qs.n_attention_wv;
-        } else if (name == LLM_TN(model.arch)(LLM_TENSOR_OUTPUT, "weight")) {
-            qs.has_output = true;
-        }
-    }
-
-    qs.n_ffn_down = qs.n_ffn_gate = qs.n_ffn_up = (int)model.hparams.n_layer;
-
-    // sanity checks
-    {
-        const auto & n_head_kv_iter = model.hparams.n_head_kv_arr.begin();
-        // attention layers have a non-zero number of kv heads
-        int32_t n_attn_layer = model.hparams.n_layer - std::count(n_head_kv_iter, n_head_kv_iter + model.hparams.n_layer, 0);
-        if (llama_model_has_encoder(&model)) {
-            n_attn_layer *= 3;
-        }
-        GGML_ASSERT((qs.n_attention_wv == n_attn_layer) && "n_attention_wv is unexpected");
-    }
-
-    size_t total_size_org = 0;
-    size_t total_size_new = 0;
-
-    std::vector workers;
-    workers.reserve(nthread);
-
-    int idx = 0;
-
-    std::vector> read_data;
-    std::vector> work;
-    std::vector> f32_conv_buf;
-
-    uint16_t n_split = 1;
-
-    // Assume split index is continuous
-    if (params->keep_split) {
-        for (const auto * it : tensors) {
-            n_split = std::max(uint16_t(it->idx + 1), n_split);
-        }
-    }
-    std::vector ctx_outs(n_split);
-    ctx_outs[0] = std::move(ctx_out);
-
-    // populate the original tensors so we get an initial meta data
-    for (const auto * it : tensors) {
-        uint16_t i_split = params->keep_split ? it->idx : 0;
-        struct ggml_tensor * tensor = it->tensor;
-        if (!ctx_outs[i_split]) {
-            ctx_outs[i_split].reset(gguf_init_empty());
-        }
-        gguf_add_tensor(ctx_outs[i_split].get(), tensor);
-    }
-
-    // Set split info if needed
-    if (n_split > 1) {
-        for (size_t i = 0; i < ctx_outs.size(); ++i) {
-            gguf_set_val_u16(ctx_outs[i].get(), ml.llm_kv(LLM_KV_SPLIT_NO).c_str(), i);
-            gguf_set_val_u16(ctx_outs[i].get(), ml.llm_kv(LLM_KV_SPLIT_COUNT).c_str(), n_split);
-            gguf_set_val_i32(ctx_outs[i].get(), ml.llm_kv(LLM_KV_SPLIT_TENSORS_COUNT).c_str(), ml.n_tensors);
-        }
-    }
-
-    int cur_split = -1;
-    std::ofstream fout;
-    auto close_ofstream = [&]() {
-        // Write metadata and close file handler
-        if (fout.is_open()) {
-            fout.seekp(0);
-            std::vector data(gguf_get_meta_size(ctx_outs[cur_split].get()));
-            gguf_get_meta_data(ctx_outs[cur_split].get(), data.data());
-            fout.write((const char *) data.data(), data.size());
-            fout.close();
-        }
-    };
-    auto new_ofstream = [&](int index) {
-        cur_split = index;
-        GGML_ASSERT(ctx_outs[cur_split] && "Find uninitialized gguf_context");
-        std::string fname = fname_out;
-        if (params->keep_split) {
-            char split_path[PATH_MAX] = {0};
-            llama_split_path(split_path, sizeof(split_path), fname_out.c_str(), cur_split, n_split);
-            fname = std::string(split_path);
-        }
-
-        fout = std::ofstream(fname, std::ios::binary);
-        fout.exceptions(std::ofstream::failbit); // fail fast on write errors
-        const size_t meta_size = gguf_get_meta_size(ctx_outs[cur_split].get());
-        // placeholder for the meta data
-        ::zeros(fout, meta_size);
-    };
-
-    const auto tn = LLM_TN(model.arch);
-    new_ofstream(0);
-    for (const auto * it : tensors) {
-        const auto & weight = *it;
-        struct ggml_tensor * tensor = weight.tensor;
-        if (weight.idx != cur_split && params->keep_split) {
-            close_ofstream();
-            new_ofstream(weight.idx);
-        }
-
-        const std::string name = ggml_get_name(tensor);
-
-        if (!ml.use_mmap) {
-            if (read_data.size() < ggml_nbytes(tensor)) {
-                read_data.resize(ggml_nbytes(tensor));
-            }
-            tensor->data = read_data.data();
-        }
-        ml.load_data_for(tensor);
-
-        LLAMA_LOG_INFO("[%4d/%4d] %36s - [%s], type = %6s, ",
-               ++idx, ml.n_tensors,
-               ggml_get_name(tensor),
-               llama_format_tensor_shape(tensor).c_str(),
-               ggml_type_name(tensor->type));
-
-        // This used to be a regex, but  has an extreme cost to compile times.
-        bool quantize = name.rfind("weight") == name.size() - 6; // ends with 'weight'?
-
-        // quantize only 2D and 3D tensors (experts)
-        quantize &= (ggml_n_dims(tensor) >= 2);
-
-        // do not quantize norm tensors
-        quantize &= name.find("_norm.weight") == std::string::npos;
-
-        quantize &= params->quantize_output_tensor || name != "output.weight";
-        quantize &= !params->only_copy;
-
-        // do not quantize expert gating tensors
-        // NOTE: can't use LLM_TN here because the layer number is not known
-        quantize &= name.find("ffn_gate_inp.weight") == std::string::npos;
-
-        // do not quantize positional embeddings and token types (BERT)
-        quantize &= name != LLM_TN(model.arch)(LLM_TENSOR_POS_EMBD,    "weight");
-        quantize &= name != LLM_TN(model.arch)(LLM_TENSOR_TOKEN_TYPES, "weight");
-
-        // do not quantize Mamba's small yet 2D weights
-        // NOTE: can't use LLM_TN here because the layer number is not known
-        quantize &= name.find("ssm_conv1d.weight") == std::string::npos;
-
-        // do not quantize RWKV's time_mix_first tensors
-        quantize &= name.find("time_mix_first.weight") == std::string::npos;
-        quantize &= name.find("time_mix_w1.weight") == std::string::npos;
-        quantize &= name.find("time_mix_w2.weight") == std::string::npos;
-        quantize &= name.find("time_mix_decay_w1.weight") == std::string::npos;
-        quantize &= name.find("time_mix_decay_w2.weight") == std::string::npos;
-
-        // do not quantize relative position bias (T5)
-        quantize &= name.find("attn_rel_b.weight") == std::string::npos;
-
-        enum ggml_type new_type;
-        void * new_data;
-        size_t new_size;
-
-        if (quantize) {
-            new_type = default_type;
-
-            // get more optimal quantization type based on the tensor shape, layer, etc.
-            if (!params->pure && ggml_is_quantized(default_type)) {
-                new_type = llama_tensor_get_type(qs, new_type, tensor, ftype);
-            }
-            if (params->token_embedding_type < GGML_TYPE_COUNT && strcmp(tensor->name, "token_embd.weight") == 0) {
-                new_type = params->token_embedding_type;
-            }
-            if (params->output_tensor_type < GGML_TYPE_COUNT && strcmp(tensor->name, "output.weight") == 0) {
-                new_type = params->output_tensor_type;
-            }
-
-            // If we've decided to quantize to the same type the tensor is already
-            // in then there's nothing to do.
-            quantize = tensor->type != new_type;
-        }
-
-        if (!quantize) {
-            new_type = tensor->type;
-            new_data = tensor->data;
-            new_size = ggml_nbytes(tensor);
-            LLAMA_LOG_INFO("size = %8.3f MB\n", ggml_nbytes(tensor)/1024.0/1024.0);
-        } else {
-            const int64_t nelements = ggml_nelements(tensor);
-
-            const float * imatrix = nullptr;
-            if (imatrix_data) {
-                auto it = imatrix_data->find(tensor->name);
-                if (it == imatrix_data->end()) {
-                    LLAMA_LOG_INFO("\n====== %s: did not find weights for %s\n", __func__, tensor->name);
-                } else {
-                    if (it->second.size() == (size_t)tensor->ne[0]*tensor->ne[2]) {
-                        imatrix = it->second.data();
-                    } else {
-                        LLAMA_LOG_INFO("\n====== %s: imatrix size %d is different from tensor size %d for %s\n", __func__,
-                                int(it->second.size()), int(tensor->ne[0]*tensor->ne[2]), tensor->name);
-
-                        // this can happen when quantizing an old mixtral model with split tensors with a new incompatible imatrix
-                        // this is a significant error and it may be good idea to abort the process if this happens,
-                        // since many people will miss the error and not realize that most of the model is being quantized without an imatrix
-                        // tok_embd should be ignored in this case, since it always causes this warning
-                        if (name != tn(LLM_TENSOR_TOKEN_EMBD, "weight")) {
-                            throw std::runtime_error(format("imatrix size %d is different from tensor size %d for %s",
-                                    int(it->second.size()), int(tensor->ne[0]*tensor->ne[2]), tensor->name));
-                        }
-                    }
-                }
-            }
-            if ((new_type == GGML_TYPE_IQ2_XXS ||
-                 new_type == GGML_TYPE_IQ2_XS  ||
-                 new_type == GGML_TYPE_IQ2_S   ||
-                 new_type == GGML_TYPE_IQ1_S   ||
-                (new_type == GGML_TYPE_IQ1_M && strcmp(tensor->name, "token_embd.weight") && strcmp(tensor->name, "output.weight"))  ||
-                (new_type == GGML_TYPE_Q2_K && params->ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S && strcmp(tensor->name, "token_embd.weight") != 0)) && !imatrix) {
-                LLAMA_LOG_ERROR("\n\n============================================================\n");
-                LLAMA_LOG_ERROR("Missing importance matrix for tensor %s in a very low-bit quantization\n", tensor->name);
-                LLAMA_LOG_ERROR("The result will be garbage, so bailing out\n");
-                LLAMA_LOG_ERROR("============================================================\n\n");
-                throw std::runtime_error(format("Missing importance matrix for tensor %s in a very low-bit quantization", tensor->name));
-            }
-
-            float * f32_data;
-
-            if (tensor->type == GGML_TYPE_F32) {
-                f32_data = (float *) tensor->data;
-            } else if (ggml_is_quantized(tensor->type) && !params->allow_requantize) {
-                throw std::runtime_error(format("requantizing from type %s is disabled", ggml_type_name(tensor->type)));
-            } else {
-                llama_tensor_dequantize_internal(tensor, f32_conv_buf, workers, nelements, nthread);
-                f32_data = (float *) f32_conv_buf.data();
-            }
-
-            LLAMA_LOG_INFO("converting to %s .. ", ggml_type_name(new_type));
-            fflush(stdout);
-
-            if (work.size() < (size_t)nelements * 4) {
-                work.resize(nelements * 4); // upper bound on size
-            }
-            new_data = work.data();
-
-            const int64_t n_per_row = tensor->ne[0];
-            const int64_t nrows = tensor->ne[1];
-
-            static const int64_t min_chunk_size = 32 * 512;
-            const int64_t chunk_size = (n_per_row >= min_chunk_size ? n_per_row : n_per_row * ((min_chunk_size + n_per_row - 1)/n_per_row));
-
-            const int64_t nelements_matrix = tensor->ne[0] * tensor->ne[1];
-            const int64_t nchunk = (nelements_matrix + chunk_size - 1)/chunk_size;
-            const int64_t nthread_use = nthread > 1 ? std::max((int64_t)1, std::min((int64_t)nthread, nchunk)) : 1;
-
-            // quantize each expert separately since they have different importance matrices
-            new_size = 0;
-            for (int64_t i03 = 0; i03 < tensor->ne[2]; ++i03) {
-                const float * f32_data_03 = f32_data + i03 * nelements_matrix;
-                void * new_data_03 = (char *)new_data + ggml_row_size(new_type, n_per_row) * i03 * nrows;
-                const float * imatrix_03 = imatrix ? imatrix + i03 * n_per_row : nullptr;
-
-                new_size += llama_tensor_quantize_internal(new_type, f32_data_03, new_data_03, chunk_size, nrows, n_per_row, imatrix_03, workers, nthread_use);
-            }
-            LLAMA_LOG_INFO("size = %8.2f MiB -> %8.2f MiB\n", ggml_nbytes(tensor)/1024.0/1024.0, new_size/1024.0/1024.0);
-        }
-        total_size_org += ggml_nbytes(tensor);
-        total_size_new += new_size;
-
-        // update the gguf meta data as we go
-        gguf_set_tensor_type(ctx_outs[cur_split].get(), name.c_str(), new_type);
-        gguf_set_tensor_data(ctx_outs[cur_split].get(), name.c_str(), new_data, new_size);
-
-        // write tensor data + padding
-        fout.write((const char *) new_data, new_size);
-        zeros(fout, GGML_PAD(new_size, align) - new_size);
-    }
-    close_ofstream();
-
-    LLAMA_LOG_INFO("%s: model size  = %8.2f MB\n", __func__, total_size_org/1024.0/1024.0);
-    LLAMA_LOG_INFO("%s: quant size  = %8.2f MB\n", __func__, total_size_new/1024.0/1024.0);
-
-    if (qs.n_fallback > 0) {
-        LLAMA_LOG_WARN("%s: WARNING: %d of %d tensor(s) required fallback quantization\n",
-                __func__, qs.n_fallback, qs.n_k_quantized + qs.n_fallback);
-    }
-}
-
-static void llama_lora_adapter_init_internal(struct llama_model * model, const char * path_lora, struct llama_lora_adapter & adapter) {
-    LLAMA_LOG_INFO("%s: loading lora adapter from '%s' ...\n", __func__, path_lora);
-
-    ggml_context * ctx_init;
-    struct gguf_init_params meta_gguf_params = {
-        /* .no_alloc = */ true,
-        /* .ctx      = */ &ctx_init,
-    };
-
-    gguf_context_ptr ctx_gguf { gguf_init_from_file(path_lora, meta_gguf_params) };
-    if (!ctx_gguf) {
-        throw std::runtime_error("failed to load lora adapter file from " + std::string(path_lora));
-    }
-
-    ggml_context_ptr ctx { ctx_init };
-
-    // check metadata
-    {
-        auto get_kv_str = [&](const std::string & key) -> std::string {
-            int id = gguf_find_key(ctx_gguf.get(), key.c_str());
-            return id < 0 ? "" : std::string(gguf_get_val_str(ctx_gguf.get(), id));
-        };
-        auto get_kv_f32 = [&](const std::string & key) -> float {
-            int id = gguf_find_key(ctx_gguf.get(), key.c_str());
-            return id < 0 ? 0.0f : gguf_get_val_f32(ctx_gguf.get(), id);
-        };
-        LLM_KV llm_kv = LLM_KV(LLM_ARCH_UNKNOWN);
-
-        auto general_type = get_kv_str(llm_kv(LLM_KV_GENERAL_TYPE));
-        if (general_type != "adapter") {
-            throw std::runtime_error("expect general.type to be 'adapter', but got: " + general_type);
-        }
-
-        auto general_arch_str = get_kv_str(llm_kv(LLM_KV_GENERAL_ARCHITECTURE));
-        auto general_arch = llm_arch_from_string(general_arch_str);
-        if (general_arch != model->arch) {
-            throw std::runtime_error("model arch and LoRA arch mismatch");
-        }
-
-        auto adapter_type = get_kv_str(llm_kv(LLM_KV_ADAPTER_TYPE));
-        if (adapter_type != "lora") {
-            throw std::runtime_error("expect adapter.type to be 'lora', but got: " + adapter_type);
-        }
-
-        adapter.alpha = get_kv_f32(llm_kv(LLM_KV_ADAPTER_LORA_ALPHA));
-    }
-
-    int n_tensors = gguf_get_n_tensors(ctx_gguf.get());
-
-    // contexts for each buffer type
-    std::map ctx_map;
-    auto ctx_for_buft = [&](ggml_backend_buffer_type_t buft) -> ggml_context * {
-        auto it = ctx_map.find(buft);
-        if (it == ctx_map.end()) {
-            // add a new context
-            struct ggml_init_params params = {
-                /*.mem_size   =*/ n_tensors*ggml_tensor_overhead(),
-                /*.mem_buffer =*/ NULL,
-                /*.no_alloc   =*/ true,
-            };
-            ggml_context * buft_ctx = ggml_init(params);
-            if (!buft_ctx) {
-                return nullptr;
-            }
-            ctx_map[buft] = buft_ctx;
-            adapter.ctxs.emplace_back(buft_ctx);
-            return buft_ctx;
-        };
-        return it->second;
-    };
-
-    // bundle lora_a and lora_b into pairs
-    std::map ab_map;
-    auto str_endswith = [](const std::string & str, const std::string & suffix) {
-        return str.size() >= suffix.size() && str.compare(str.size()-suffix.size(), suffix.size(), suffix) == 0;
-    };
-    for (ggml_tensor * cur = ggml_get_first_tensor(ctx.get()); cur; cur = ggml_get_next_tensor(ctx.get(), cur)) {
-        std::string name(cur->name);
-        if (str_endswith(name, ".lora_a")) {
-            replace_all(name, ".lora_a", "");
-            if (ab_map.find(name) == ab_map.end()) {
-                ab_map[name] = llama_lora_weight(cur, nullptr);
-            } else {
-                ab_map[name].a = cur;
-            }
-        } else if (str_endswith(name, ".lora_b")) {
-            replace_all(name, ".lora_b", "");
-            if (ab_map.find(name) == ab_map.end()) {
-                ab_map[name] = llama_lora_weight(nullptr, cur);
-            } else {
-                ab_map[name].b = cur;
-            }
-        } else {
-            throw std::runtime_error("LoRA tensor '" + name + "' has unexpected suffix");
-        }
-    }
-
-    // add tensors
-    for (auto & it : ab_map) {
-        const std::string & name = it.first;
-        llama_lora_weight & w = it.second;
-
-        if (!w.a || !w.b) {
-            throw std::runtime_error("LoRA tensor pair for '" + name + "' is missing one component");
-        }
-
-        // device buft and device ctx
-        auto * model_tensor = llama_get_model_tensor(model, name.c_str());
-        if (!model_tensor) {
-            throw std::runtime_error("LoRA tensor '" + name + "' does not exist in base model");
-        }
-        struct ggml_context * dev_ctx = ctx_for_buft(ggml_backend_buffer_get_type(model_tensor->buffer));
-        // validate tensor shape
-        if (model_tensor->ne[0] != w.a->ne[0] || model_tensor->ne[1] != w.b->ne[1]) {
-            throw std::runtime_error("tensor '" + name + "' has incorrect shape");
-        }
-        if (w.a->ne[1] != w.b->ne[0]) {
-            throw std::runtime_error("lora_a tensor is not transposed (hint: adapter from \"finetune\" example is no longer supported)");
-        }
-        // save tensor to adapter
-        struct ggml_tensor * tensor_a = ggml_dup_tensor(dev_ctx, w.a);
-        struct ggml_tensor * tensor_b = ggml_dup_tensor(dev_ctx, w.b);
-        ggml_set_name(tensor_a, w.a->name);
-        ggml_set_name(tensor_b, w.b->name);
-        adapter.ab_map[name] = llama_lora_weight(tensor_a, tensor_b);
-    }
-
-    // allocate tensors / buffers and zero
-    {
-        adapter.ctxs.reserve(ctx_map.size());
-        adapter.bufs.reserve(ctx_map.size());
-        for (auto & it : ctx_map) {
-            ggml_backend_buffer_type_t buft = it.first;
-            ggml_context * ctx_dev = it.second;
-            ggml_backend_buffer_ptr buf { ggml_backend_alloc_ctx_tensors_from_buft(ctx_dev, buft) };
-            if (!buf) {
-                throw std::runtime_error("failed to allocate buffer for lora adapter\n");
-            }
-            LLAMA_LOG_INFO("%s: %10s LoRA buffer size = %8.2f MiB\n", __func__, ggml_backend_buffer_name(buf.get()), ggml_backend_buffer_get_size(buf.get())/1024.0/1024.0);
-            adapter.bufs.emplace_back(std::move(buf));
-        }
-    }
-
-    // set tensor data
-    {
-        llama_file gguf_file(path_lora, "rb");
-        std::vector read_buf;
-        auto set_tensor = [&](struct ggml_tensor * orig, struct ggml_tensor * dev) {
-            size_t offs = gguf_get_data_offset(ctx_gguf.get()) + gguf_get_tensor_offset(ctx_gguf.get(), gguf_find_tensor(ctx_gguf.get(), orig->name));
-            size_t size = ggml_nbytes(orig);
-            read_buf.resize(size);
-            gguf_file.seek(offs, SEEK_SET);
-            gguf_file.read_raw(read_buf.data(), size);
-            ggml_backend_tensor_set(dev, read_buf.data(), 0, size);
-        };
-        for (auto & it : adapter.ab_map) {
-            auto orig = ab_map[it.first];
-            auto dev  = it.second;
-            set_tensor(orig.a, dev.a);
-            set_tensor(orig.b, dev.b);
-        }
-    }
-
-    LLAMA_LOG_INFO("%s: loaded %zu tensors from lora file\n", __func__, adapter.ab_map.size()*2);
-}
-
-int32_t llama_lora_adapter_set(
-            struct llama_context * ctx,
-            struct llama_lora_adapter * adapter,
-            float scale) {
-    if (ctx->cparams.flash_attn) {
-        LLAMA_LOG_ERROR("%s: flash_attn is not compatible with LoRA\n", __func__);
-        return -1;
-    }
-    ctx->lora_adapters[adapter] = scale;
-    return 0;
-}
-
-int32_t llama_lora_adapter_remove(
-            struct llama_context * ctx,
-            struct llama_lora_adapter * adapter) {
-    auto pos = ctx->lora_adapters.find(adapter);
-    if (pos != ctx->lora_adapters.end()) {
-        ctx->lora_adapters.erase(pos);
-        return 0;
-    }
-    return -1;
-}
-
-void llama_lora_adapter_clear(struct llama_context * ctx) {
-    ctx->lora_adapters.clear();
-}
-
-void llama_lora_adapter_free(struct llama_lora_adapter * adapter) {
-    delete adapter;
-}
-
-//
-// interface implementation
-//
-struct llama_model_params llama_model_default_params() {
-    struct llama_model_params result = {
-        /*.devices                     =*/ nullptr,
-        /*.n_gpu_layers                =*/ 0,
-        /*.split_mode                  =*/ LLAMA_SPLIT_MODE_LAYER,
-        /*.main_gpu                    =*/ 0,
-        /*.tensor_split                =*/ nullptr,
-        /*.rpc_servers                 =*/ nullptr,
-        /*.progress_callback           =*/ nullptr,
-        /*.progress_callback_user_data =*/ nullptr,
-        /*.kv_overrides                =*/ nullptr,
-        /*.vocab_only                  =*/ false,
-        /*.use_mmap                    =*/ true,
-        /*.use_mlock                   =*/ false,
-        /*.check_tensors               =*/ false,
-    };
-
-#ifdef GGML_USE_METAL
-    // note: we usually have plenty of VRAM, so by default offload all layers to the GPU
-    result.n_gpu_layers = 999;
-#endif
-
-    return result;
-}
-
-struct llama_context_params llama_context_default_params() {
-    struct llama_context_params result = {
-        /*.n_ctx                       =*/ 512,
-        /*.n_batch                     =*/ 2048,
-        /*.n_ubatch                    =*/ 512,
-        /*.n_seq_max                   =*/ 1,
-        /*.n_threads                   =*/ GGML_DEFAULT_N_THREADS, // TODO: better default
-        /*.n_threads_batch             =*/ GGML_DEFAULT_N_THREADS,
-        /*.rope_scaling_type           =*/ LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED,
-        /*.pooling_type                =*/ LLAMA_POOLING_TYPE_UNSPECIFIED,
-        /*.attention_type              =*/ LLAMA_ATTENTION_TYPE_UNSPECIFIED,
-        /*.rope_freq_base              =*/ 0.0f,
-        /*.rope_freq_scale             =*/ 0.0f,
-        /*.yarn_ext_factor             =*/ -1.0f,
-        /*.yarn_attn_factor            =*/ 1.0f,
-        /*.yarn_beta_fast              =*/ 32.0f,
-        /*.yarn_beta_slow              =*/ 1.0f,
-        /*.yarn_orig_ctx               =*/ 0,
-        /*.defrag_thold                =*/ -1.0f,
-        /*.cb_eval                     =*/ nullptr,
-        /*.cb_eval_user_data           =*/ nullptr,
-        /*.type_k                      =*/ GGML_TYPE_F16,
-        /*.type_v                      =*/ GGML_TYPE_F16,
-        /*.logits_all                  =*/ false,
-        /*.embeddings                  =*/ false,
-        /*.offload_kqv                 =*/ true,
-        /*.flash_attn                  =*/ false,
-        /*.no_perf                     =*/ true,
-        /*.abort_callback              =*/ nullptr,
-        /*.abort_callback_data         =*/ nullptr,
-    };
-
-    return result;
-}
-
-struct llama_sampler_chain_params llama_sampler_chain_default_params() {
-    struct llama_sampler_chain_params result = {
-        /*.no_perf                     =*/ true,
-    };
-
-    return result;
-}
-
-struct llama_model_quantize_params llama_model_quantize_default_params() {
-    struct llama_model_quantize_params result = {
-        /*.nthread                     =*/ 0,
-        /*.ftype                       =*/ LLAMA_FTYPE_MOSTLY_Q5_1,
-        /*.output_tensor_type          =*/ GGML_TYPE_COUNT,
-        /*.token_embedding_type        =*/ GGML_TYPE_COUNT,
-        /*.allow_requantize            =*/ false,
-        /*.quantize_output_tensor      =*/ true,
-        /*.only_copy                   =*/ false,
-        /*.pure                        =*/ false,
-        /*.keep_split                  =*/ false,
-        /*.imatrix                     =*/ nullptr,
-        /*.kv_overrides                =*/ nullptr,
-    };
-
-    return result;
-}
-
-size_t llama_max_devices(void) {
-    return 16;
-}
-
-bool llama_supports_mmap(void) {
-    return llama_mmap::SUPPORTED;
-}
-
-bool llama_supports_mlock(void) {
-    return llama_mlock::SUPPORTED;
-}
-
-bool llama_supports_gpu_offload(void) {
-    return ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_GPU) != nullptr ||
-           llama_supports_rpc();
-}
-
-bool llama_supports_rpc(void) {
-    return ggml_backend_reg_by_name("RPC") != nullptr;
-}
-
-void llama_backend_init(void) {
-    ggml_time_init();
-
-    // needed to initialize f16 tables
-    {
-        struct ggml_init_params params = { 0, NULL, false };
-        struct ggml_context * ctx = ggml_init(params);
-        ggml_free(ctx);
-    }
-}
-
-void llama_numa_init(enum ggml_numa_strategy numa) {
-    if (numa != GGML_NUMA_STRATEGY_DISABLED) {
-        auto * dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
-        GGML_ASSERT(dev && "CPU backend is not loaded");
-        auto * reg = ggml_backend_dev_backend_reg(dev);
-        auto * numa_init_fn = (decltype(ggml_numa_init) *) ggml_backend_reg_get_proc_address(reg, "ggml_backend_cpu_numa_init");
-        numa_init_fn(numa);
-    }
-}
-
-void llama_attach_threadpool(
-             struct llama_context * ctx,
-        ggml_threadpool_t   threadpool,
-        ggml_threadpool_t   threadpool_batch) {
-    ctx->threadpool       = threadpool;
-    ctx->threadpool_batch = threadpool_batch ? threadpool_batch : threadpool;
-}
-
-void llama_detach_threadpool(struct llama_context * ctx) {
-    ctx->threadpool       = nullptr;
-    ctx->threadpool_batch = nullptr;
-}
-
-void llama_backend_free(void) {
-    ggml_quantize_free();
-}
-
-int64_t llama_time_us(void) {
-    return ggml_time_us();
+    return result;
 }
 
-struct llama_model * llama_load_model_from_file(
-        const char * path_model,
-        struct llama_model_params   params) {
-    ggml_time_init();
-
-    llama_model * model = new llama_model;
-
-    unsigned cur_percentage = 0;
-    if (params.progress_callback == NULL) {
-        params.progress_callback_user_data = &cur_percentage;
-        params.progress_callback = [](float progress, void * ctx) {
-            unsigned * cur_percentage_p = (unsigned *) ctx;
-            unsigned percentage = (unsigned) (100 * progress);
-            while (percentage > *cur_percentage_p) {
-                *cur_percentage_p = percentage;
-                LLAMA_LOG_CONT(".");
-                if (percentage >= 100) {
-                    LLAMA_LOG_CONT("\n");
-                }
-            }
-            return true;
-        };
-    }
-
-    if (params.rpc_servers != nullptr && params.rpc_servers[0] != '\0') {
-        // split the servers set them into model->rpc_servers
-        std::string servers(params.rpc_servers);
-        size_t pos = 0;
-        while ((pos = servers.find(',')) != std::string::npos) {
-            std::string server = servers.substr(0, pos);
-            model->rpc_servers.push_back(server);
-            servers.erase(0, pos + 1);
-        }
-        model->rpc_servers.push_back(servers);
-    }
+static struct ggml_cgraph * llama_build_graph(
+         llama_context & lctx,
+    const llama_ubatch & ubatch,
+                  bool   worst_case) {
+    const auto & model = lctx.model;
 
-    // add RPC devices
-    if (!model->rpc_servers.empty()) {
-        ggml_backend_reg_t rpc_reg = ggml_backend_reg_by_name("RPC");
-        if (!rpc_reg) {
-            LLAMA_LOG_ERROR("%s: failed to find RPC backend\n", __func__);
-            llama_free_model(model);
-            return nullptr;
+    // this callback allows us to apply custom logic to each tensor (e.g. ggml-alloc, offloading, etc.)
+    llm_build_cb cb = [&](struct ggml_tensor * cur, const char * name, int il) {
+        if (il >= 0) {
+            ggml_format_name(cur, "%s-%d", name, il);
+        } else {
+            ggml_set_name(cur, name);
         }
 
-        typedef ggml_backend_dev_t (*ggml_backend_rpc_add_device_t)(const char * endpoint);
-        ggml_backend_rpc_add_device_t ggml_backend_rpc_add_device_fn = (ggml_backend_rpc_add_device_t) ggml_backend_reg_get_proc_address(rpc_reg, "ggml_backend_rpc_add_device");
-        if (!ggml_backend_rpc_add_device_fn) {
-            LLAMA_LOG_ERROR("%s: failed to find RPC device add function\n", __func__);
-            llama_free_model(model);
-            return nullptr;
+        if (!lctx.cparams.offload_kqv) {
+            if (strcmp(name, "kqv_merged_cont") == 0) {
+                // all nodes between the KV store and the attention output are run on the CPU
+                ggml_backend_sched_set_tensor_backend(lctx.sched.get(), cur, lctx.backend_cpu);
+            }
         }
 
-        for (const std::string & server : model->rpc_servers) {
-            ggml_backend_dev_t dev = ggml_backend_rpc_add_device_fn(server.c_str());
-            if (dev) {
-                model->devices.push_back(dev);
-            } else {
-                LLAMA_LOG_ERROR("%s: failed to add RPC device for server '%s'\n", __func__, server.c_str());
-                llama_free_model(model);
-                return nullptr;
+        // norm may be automatically assigned to the backend of the previous layer, increasing data transfer between backends
+        // FIXME: fix in ggml_backend_sched
+        const bool full_offload = lctx.model.n_gpu_layers > (int)lctx.model.hparams.n_layer;
+        if (ubatch.n_tokens < 32 || full_offload) {
+            if (il != -1 && strcmp(name, "norm") == 0) {
+                const auto & dev_layer = lctx.model.dev_layer.at(il);
+                for (auto & backend : lctx.backends) {
+                    if (ggml_backend_get_device(backend.get()) == dev_layer.dev) {
+                        if (ggml_backend_supports_op(backend.get(), cur)) {
+                            ggml_backend_sched_set_tensor_backend(lctx.sched.get(), cur, backend.get());
+                        }
+                    }
+                }
             }
         }
-    }
+    };
 
-    // create list of devices to use with this model
-    if (params.devices) {
-        for (ggml_backend_dev_t * dev = params.devices; *dev; ++dev) {
-            model->devices.push_back(*dev);
-        }
-    } else {
-        // use all available devices
-        for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
-            ggml_backend_dev_t dev = ggml_backend_dev_get(i);
-            switch (ggml_backend_dev_type(dev)) {
-                case GGML_BACKEND_DEVICE_TYPE_CPU:
-                case GGML_BACKEND_DEVICE_TYPE_ACCEL:
-                    // skip CPU backends since they are handled separately
-                    break;
+    struct ggml_cgraph * result = NULL;
 
-                case GGML_BACKEND_DEVICE_TYPE_GPU:
-                    model->devices.push_back(dev);
-                    break;
-            }
-        }
-    }
+    struct llm_build_context llm(lctx, ubatch, cb, worst_case);
 
-    // if using single GPU mode, remove all except the main GPU
-    if (params.split_mode == LLAMA_SPLIT_MODE_NONE) {
-        if (params.main_gpu < 0 || params.main_gpu >= (int)model->devices.size()) {
-            LLAMA_LOG_ERROR("%s: invalid value for main_gpu: %d (available devices: %d)\n", __func__, params.main_gpu, (int)model->devices.size());
-            llama_free_model(model);
-            return nullptr;
-        }
-        ggml_backend_dev_t main_gpu = model->devices[params.main_gpu];
-        model->devices.clear();
-        model->devices.push_back(main_gpu);
-    }
+    llm.init();
 
-    for (auto * dev : model->devices) {
-        size_t free, total; // NOLINT
-        ggml_backend_dev_memory(dev, &free, &total);
-        LLAMA_LOG_INFO("%s: using device %s (%s) - %zu MiB free\n", __func__, ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), free/1024/1024);
+    switch (model.arch) {
+        case LLM_ARCH_LLAMA:
+        case LLM_ARCH_MINICPM:
+        case LLM_ARCH_GRANITE:
+        case LLM_ARCH_GRANITE_MOE:
+            {
+                result = llm.build_llama();
+            } break;
+        case LLM_ARCH_DECI:
+            {
+                result = llm.build_deci();
+            } break;
+        case LLM_ARCH_BAICHUAN:
+            {
+                result = llm.build_baichuan();
+            } break;
+        case LLM_ARCH_FALCON:
+            {
+                result = llm.build_falcon();
+            } break;
+        case LLM_ARCH_GROK:
+            {
+                result = llm.build_grok();
+            } break;
+        case LLM_ARCH_STARCODER:
+            {
+                result = llm.build_starcoder();
+            } break;
+        case LLM_ARCH_REFACT:
+            {
+                result = llm.build_refact();
+            } break;
+        case LLM_ARCH_BERT:
+        case LLM_ARCH_JINA_BERT_V2:
+        case LLM_ARCH_NOMIC_BERT:
+            {
+                result = llm.build_bert();
+            } break;
+        case LLM_ARCH_BLOOM:
+            {
+                result = llm.build_bloom();
+            } break;
+        case LLM_ARCH_MPT:
+            {
+                result = llm.build_mpt();
+            } break;
+         case LLM_ARCH_STABLELM:
+            {
+                result = llm.build_stablelm();
+            } break;
+        case LLM_ARCH_QWEN:
+            {
+                result = llm.build_qwen();
+            } break;
+        case LLM_ARCH_QWEN2:
+            {
+                result = llm.build_qwen2();
+            } break;
+        case LLM_ARCH_QWEN2VL:
+            {
+                lctx.n_pos_per_token = 4;
+                result = llm.build_qwen2vl();
+            } break;
+        case LLM_ARCH_QWEN2MOE:
+            {
+                result = llm.build_qwen2moe();
+            } break;
+        case LLM_ARCH_PHI2:
+            {
+                result = llm.build_phi2();
+            } break;
+        case LLM_ARCH_PHI3:
+            {
+                result = llm.build_phi3();
+            } break;
+        case LLM_ARCH_PLAMO:
+            {
+                result = llm.build_plamo();
+            } break;
+        case LLM_ARCH_GPT2:
+            {
+                result = llm.build_gpt2();
+            } break;
+        case LLM_ARCH_CODESHELL:
+            {
+                result = llm.build_codeshell();
+            } break;
+        case LLM_ARCH_ORION:
+            {
+                result = llm.build_orion();
+            } break;
+        case LLM_ARCH_INTERNLM2:
+            {
+                result = llm.build_internlm2();
+            } break;
+        case LLM_ARCH_MINICPM3:
+            {
+                result = llm.build_minicpm3();
+            } break;
+        case LLM_ARCH_GEMMA:
+            {
+                result = llm.build_gemma();
+            } break;
+        case LLM_ARCH_GEMMA2:
+            {
+                result = llm.build_gemma2();
+            } break;
+        case LLM_ARCH_STARCODER2:
+            {
+                result = llm.build_starcoder2();
+            } break;
+        case LLM_ARCH_MAMBA:
+            {
+                result = llm.build_mamba();
+            } break;
+        case LLM_ARCH_XVERSE:
+            {
+                result = llm.build_xverse();
+            } break;
+        case LLM_ARCH_COMMAND_R:
+            {
+                result = llm.build_command_r();
+            } break;
+        case LLM_ARCH_COHERE2:
+            {
+                result = llm.build_cohere2();
+            } break;
+        case LLM_ARCH_DBRX:
+            {
+                result = llm.build_dbrx();
+            } break;
+        case LLM_ARCH_OLMO:
+            {
+                result = llm.build_olmo();
+            } break;
+        case LLM_ARCH_OLMO2:
+            {
+                result = llm.build_olmo2();
+            } break;
+        case LLM_ARCH_OLMOE:
+            {
+                result = llm.build_olmoe();
+            } break;
+        case LLM_ARCH_OPENELM:
+            {
+                result = llm.build_openelm();
+            } break;
+        case LLM_ARCH_GPTNEOX:
+            {
+                result = llm.build_gptneox();
+            } break;
+        case LLM_ARCH_ARCTIC:
+            {
+                result = llm.build_arctic();
+            } break;
+        case LLM_ARCH_DEEPSEEK:
+            {
+                result = llm.build_deepseek();
+            } break;
+        case LLM_ARCH_DEEPSEEK2:
+            {
+                result = llm.build_deepseek2();
+            } break;
+        case LLM_ARCH_CHATGLM:
+            {
+                result = llm.build_chatglm();
+            } break;
+        case LLM_ARCH_BITNET:
+            {
+                result = llm.build_bitnet();
+            } break;
+        case LLM_ARCH_T5:
+            {
+                if (lctx.is_encoding) {
+                    result = llm.build_t5_enc();
+                } else {
+                    result = llm.build_t5_dec();
+                }
+            } break;
+        case LLM_ARCH_T5ENCODER:
+            {
+                result = llm.build_t5_enc();
+            } break;
+        case LLM_ARCH_JAIS:
+            {
+                result = llm.build_jais();
+            } break;
+        case LLM_ARCH_NEMOTRON:
+            {
+                result = llm.build_nemotron();
+            } break;
+        case LLM_ARCH_EXAONE:
+            {
+                result = llm.build_exaone();
+            } break;
+        case LLM_ARCH_RWKV6:
+            {
+                result = llm.build_rwkv6();
+            } break;
+        case LLM_ARCH_CHAMELEON:
+            {
+                result = llm.build_chameleon();
+            } break;
+        case LLM_ARCH_WAVTOKENIZER_DEC:
+            {
+                result = llm.build_wavtokenizer_dec();
+            } break;
+        default:
+            GGML_ABORT("fatal error");
     }
 
-    int status = llama_model_load(path_model, *model, params);
-    GGML_ASSERT(status <= 0);
-    if (status < 0) {
-        if (status == -1) {
-            LLAMA_LOG_ERROR("%s: failed to load model\n", __func__);
-        } else if (status == -2) {
-            LLAMA_LOG_INFO("%s: cancelled model load\n", __func__);
-        }
-        llama_free_model(model);
-        return nullptr;
+    // add on pooling layer
+    if (lctx.cparams.embeddings) {
+        result = llm.append_pooling(result);
     }
 
-    return model;
-}
+    llm.free();
 
-void llama_free_model(struct llama_model * model) {
-    delete model;
+    return result;
 }
 
-struct llama_context * llama_new_context_with_model(
-                 struct llama_model * model,
-        struct llama_context_params   params) {
-
-    if (!model) {
-        LLAMA_LOG_ERROR("%s: model cannot be NULL\n", __func__);
-        return nullptr;
-    }
-
-    if (params.n_batch == 0 && params.n_ubatch == 0) {
-        LLAMA_LOG_ERROR("%s: n_batch and n_ubatch cannot both be zero\n", __func__);
-        return nullptr;
-    }
-
-    if (params.n_ctx == 0 && model->hparams.n_ctx_train == 0) {
-        LLAMA_LOG_ERROR("%s: n_ctx and model->hparams.n_ctx_train cannot both be zero\n", __func__);
-        return nullptr;
-    }
-
-    if (params.flash_attn && model->arch == LLM_ARCH_GROK) {
-        LLAMA_LOG_WARN("%s: flash_attn is not compatible with Grok - forcing off\n", __func__);
-        params.flash_attn = false;
+// returns the result of ggml_backend_sched_graph_compute_async execution
+static enum ggml_status llama_graph_compute(
+          llama_context & lctx,
+            ggml_cgraph * gf,
+                    int   n_threads,
+        ggml_threadpool * threadpool) {
+    if (lctx.backend_cpu != nullptr) {
+        auto * reg = ggml_backend_dev_backend_reg(ggml_backend_get_device(lctx.backend_cpu));
+        auto * set_threadpool_fn = (decltype(ggml_backend_cpu_set_threadpool) *) ggml_backend_reg_get_proc_address(reg, "ggml_backend_cpu_set_threadpool");
+        set_threadpool_fn(lctx.backend_cpu, threadpool);
     }
 
-    if (params.flash_attn && model->hparams.n_embd_head_k != model->hparams.n_embd_head_v) {
-        LLAMA_LOG_WARN("%s: flash_attn requires n_embd_head_k == n_embd_head_v - forcing off\n", __func__);
-        params.flash_attn = false;
+    // set the number of threads for all the backends
+    for (const auto & set_n_threads_fn : lctx.set_n_threads_fns) {
+        set_n_threads_fn.second(set_n_threads_fn.first, n_threads);
     }
 
-    if (ggml_is_quantized(params.type_v) && !params.flash_attn) {
-        LLAMA_LOG_ERROR("%s: V cache quantization requires flash_attn\n", __func__);
-        return nullptr;
+    auto status = ggml_backend_sched_graph_compute_async(lctx.sched.get(), gf);
+    if (status != GGML_STATUS_SUCCESS) {
+        LLAMA_LOG_ERROR("%s: ggml_backend_sched_graph_compute_async failed with error %d\n", __func__, status);
     }
 
-    llama_context * ctx = new llama_context(*model);
-
-    const auto & hparams = model->hparams;
-    auto       & cparams = ctx->cparams;
-
-    cparams.n_seq_max        = std::max(1u, params.n_seq_max);
-    cparams.n_threads        = params.n_threads;
-    cparams.n_threads_batch  = params.n_threads_batch;
-    cparams.yarn_ext_factor  = params.yarn_ext_factor;
-    cparams.yarn_attn_factor = params.yarn_attn_factor;
-    cparams.yarn_beta_fast   = params.yarn_beta_fast;
-    cparams.yarn_beta_slow   = params.yarn_beta_slow;
-    cparams.defrag_thold     = params.defrag_thold;
-    cparams.embeddings       = params.embeddings;
-    cparams.offload_kqv      = params.offload_kqv;
-    cparams.flash_attn       = params.flash_attn;
-    cparams.no_perf          = params.no_perf;
-    cparams.pooling_type     = params.pooling_type;
+    // fprintf(stderr, "splits: %d\n", ggml_backend_sched_get_n_splits(lctx.sched));
 
-    cparams.n_ctx            = params.n_ctx           == 0    ? hparams.n_ctx_train           : params.n_ctx;
-    cparams.rope_freq_base   = params.rope_freq_base  == 0.0f ? hparams.rope_freq_base_train  : params.rope_freq_base;
-    cparams.rope_freq_scale  = params.rope_freq_scale == 0.0f ? hparams.rope_freq_scale_train : params.rope_freq_scale;
+    return status;
+}
 
-    // this is necessary due to kv_self.n being padded later during inference
-    cparams.n_ctx            = GGML_PAD(cparams.n_ctx, llama_kv_cache_get_padding(cparams));
+// decode a batch of tokens by evaluating the transformer
+// in case of unsuccessful decoding (error or warning),
+// the kv_cache state will be returned to its original state
+// (for non-recurrent models) or cleaned (for recurrent models)
+//
+//   - lctx:      llama context
+//   - batch:     batch to evaluate
+//
+// return 0 on success
+// return positive int on warning
+// return negative int on error
+//
+static int llama_decode_impl(
+         llama_context & lctx,
+           llama_batch   inp_batch) {
 
-    // with causal attention, the batch size is limited by the context size
-    cparams.n_batch          = hparams.causal_attn ? std::min(cparams.n_ctx, params.n_batch) : params.n_batch;
+    lctx.is_encoding = false;
 
-    // the batch has to be at least GGML_KQ_MASK_PAD because we will be padding the KQ_mask
-    // this is required by GPU kernels in order to avoid out-of-bounds accesses (e.g. ggml_flash_attn_ext)
-    // ref: https://github.com/ggerganov/llama.cpp/pull/5021
-    if (cparams.n_batch < GGML_KQ_MASK_PAD) {
-        LLAMA_LOG_WARN("%s: n_batch is less than GGML_KQ_MASK_PAD - increasing to %d\n", __func__, GGML_KQ_MASK_PAD);
-        cparams.n_batch = GGML_KQ_MASK_PAD;
+    if (inp_batch.n_tokens == 0) {
+        LLAMA_LOG_ERROR("%s: n_tokens == 0\n", __func__);
+        return -1;
     }
 
-    cparams.n_ubatch         = std::min(cparams.n_batch, params.n_ubatch == 0 ? params.n_batch : params.n_ubatch);
-
-    cparams.n_ctx_orig_yarn  = params.yarn_orig_ctx    != 0 ? params.yarn_orig_ctx    :
-                               hparams.n_ctx_orig_yarn != 0 ? hparams.n_ctx_orig_yarn :
-                                                              hparams.n_ctx_train;
-
-    cparams.cb_eval           = params.cb_eval;
-    cparams.cb_eval_user_data = params.cb_eval_user_data;
-
-    auto rope_scaling_type = params.rope_scaling_type;
-    if (rope_scaling_type == LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED) {
-        rope_scaling_type = hparams.rope_scaling_type_train;
-    }
+    // temporary allocate memory for the input batch if needed
+    llama_batch_allocr batch_allocr(inp_batch, inp_batch.pos ? -1 : lctx.kv_self.max_pos() + 1);
 
-    if (rope_scaling_type == LLAMA_ROPE_SCALING_TYPE_NONE) {
-        cparams.rope_freq_scale = 1.0f; // never scale if scaling type is none
-    }
+    const llama_batch & batch = batch_allocr.batch;
+    const uint32_t n_tokens_all = batch.n_tokens;
 
-    if (cparams.yarn_ext_factor < 0.0f) { // negative indicates 'not set'
-        cparams.yarn_ext_factor = rope_scaling_type == LLAMA_ROPE_SCALING_TYPE_YARN ? 1.0f : 0.0f;
-    }
+    const auto & model   = lctx.model;
+    const auto & hparams = model.hparams;
+    const auto & cparams = lctx.cparams;
 
-    cparams.yarn_attn_factor *= hparams.rope_attn_factor;
+    GGML_ASSERT((!batch.token && batch.embd) || (batch.token && !batch.embd)); // NOLINT
 
-    if (cparams.pooling_type == LLAMA_POOLING_TYPE_UNSPECIFIED) {
-        if (hparams.pooling_type == LLAMA_POOLING_TYPE_UNSPECIFIED) {
-            cparams.pooling_type = LLAMA_POOLING_TYPE_NONE;
-        } else {
-            cparams.pooling_type = hparams.pooling_type;
+    if (batch.token) {
+        for (uint32_t i = 0; i < n_tokens_all; ++i) {
+            if (batch.token[i] < 0 || (uint32_t)batch.token[i] >= model.vocab.n_vocab) {
+                LLAMA_LOG_ERROR("%s: invalid token[%d] = %d\n", __func__, i, batch.token[i]);
+                return -1;
+            }
         }
-    }
-
-    if (params.attention_type == LLAMA_ATTENTION_TYPE_UNSPECIFIED) {
-        cparams.causal_attn = hparams.causal_attn;
-    } else {
-        cparams.causal_attn = params.attention_type == LLAMA_ATTENTION_TYPE_CAUSAL;
-    }
-
-    const uint32_t n_ctx_per_seq = cparams.n_ctx / cparams.n_seq_max;
-
-    LLAMA_LOG_INFO("%s: n_seq_max     = %u\n",   __func__, cparams.n_seq_max);
-    LLAMA_LOG_INFO("%s: n_ctx         = %u\n",   __func__, cparams.n_ctx);
-    LLAMA_LOG_INFO("%s: n_ctx_per_seq = %u\n",   __func__, n_ctx_per_seq);
-    LLAMA_LOG_INFO("%s: n_batch       = %u\n",   __func__, cparams.n_batch);
-    LLAMA_LOG_INFO("%s: n_ubatch      = %u\n",   __func__, cparams.n_ubatch);
-    LLAMA_LOG_INFO("%s: flash_attn    = %d\n",   __func__, cparams.flash_attn);
-    LLAMA_LOG_INFO("%s: freq_base     = %.1f\n", __func__, cparams.rope_freq_base);
-    LLAMA_LOG_INFO("%s: freq_scale    = %g\n",   __func__, cparams.rope_freq_scale);
-
-    if (n_ctx_per_seq < hparams.n_ctx_train) {
-        LLAMA_LOG_WARN("%s: n_ctx_per_seq (%u) < n_ctx_train (%u) -- the full capacity of the model will not be utilized\n",
-                __func__, n_ctx_per_seq, hparams.n_ctx_train);
-    }
-
-    if (n_ctx_per_seq > hparams.n_ctx_train) {
-        LLAMA_LOG_WARN("%s: n_ctx_pre_seq (%u) > n_ctx_train (%u) -- possible training context overflow\n",
-                __func__, n_ctx_per_seq, hparams.n_ctx_train);
-    }
-
-    ctx->logits_all = params.logits_all;
+    }
 
-    // build worst-case graph for encoder if a model contains encoder
-    ctx->is_encoding = llama_model_has_encoder(model);
+    GGML_ASSERT(n_tokens_all <= cparams.n_batch);
 
-    uint32_t kv_size = cparams.n_ctx;
-    ggml_type type_k = params.type_k;
-    ggml_type type_v = params.type_v;
+    GGML_ASSERT((cparams.causal_attn || cparams.n_ubatch >= n_tokens_all) && "non-causal attention requires n_ubatch >= n_tokens");
 
-    // Mamba only needs a constant number of KV cache cells per sequence
-    if (llama_model_is_recurrent(model)) {
-        // Mamba needs at least as many KV cells as there are sequences kept at any time
-        kv_size = std::max((uint32_t) 1, params.n_seq_max);
-        // it's probably best to keep as much precision as possible for the states
-        type_k = GGML_TYPE_F32; // required by ggml_ssm_conv for Mamba's conv_states
-        type_v = GGML_TYPE_F32; // required by ggml_ssm_scan for Mamba's ssm_states
+    if (lctx.t_compute_start_us == 0) {
+        lctx.t_compute_start_us = ggml_time_us();
     }
+    lctx.n_queued_tokens += n_tokens_all;
 
-    GGML_ASSERT(hparams.n_embd_head_k % ggml_blck_size(type_k) == 0);
-    GGML_ASSERT(hparams.n_embd_head_v % ggml_blck_size(type_v) == 0);
+    auto & kv_self = lctx.kv_self;
+    llama_kv_slot_restorer kv_slot_restorer(kv_self);
 
-    if (!hparams.vocab_only) {
-        // GPU backends
-        for (auto * dev : model->devices) {
-            ggml_backend_t backend = ggml_backend_dev_init(dev, nullptr);
-            if (backend == nullptr) {
-                LLAMA_LOG_ERROR("%s: failed to initialize %s backend\n", __func__, ggml_backend_dev_name(dev));
-                llama_free(ctx);
-                return nullptr;
-            }
-            ctx->backends.emplace_back(backend);
-        }
+    const int64_t n_embd  = hparams.n_embd;
+    const int64_t n_vocab = hparams.n_vocab;
 
-        // add ACCEL backends (such as BLAS)
-        for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
-            ggml_backend_dev_t dev = ggml_backend_dev_get(i);
-            if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_ACCEL) {
-                ggml_backend_t backend = ggml_backend_dev_init(dev, nullptr);
-                if (backend == nullptr) {
-                    LLAMA_LOG_ERROR("%s: failed to initialize %s backend\n", __func__, ggml_backend_dev_name(dev));
-                    llama_free(ctx);
-                    return nullptr;
-                }
-                ctx->backends.emplace_back(backend);
-            }
-        }
+    uint32_t n_outputs = 0;
+    uint32_t n_outputs_prev = 0;
 
-        // add CPU backend
-        ctx->backend_cpu = ggml_backend_init_by_type(GGML_BACKEND_DEVICE_TYPE_CPU, nullptr);
-        if (ctx->backend_cpu == nullptr) {
-            LLAMA_LOG_ERROR("%s: failed to initialize CPU backend\n", __func__);
-            llama_free(ctx);
-            return nullptr;
-        }
-        ctx->backends.emplace_back(ctx->backend_cpu);
+    const auto n_ubatch = cparams.n_ubatch;
 
-        // create a list of the set_n_threads functions in the backends
-        for (auto & backend : ctx->backends) {
-            ggml_backend_dev_t dev = ggml_backend_get_device(backend.get());
-            ggml_backend_reg_t reg = dev ? ggml_backend_dev_backend_reg(dev) : nullptr;
-            if (reg) {
-                auto ggml_backend_set_n_threads_fn = (ggml_backend_set_n_threads_t) ggml_backend_reg_get_proc_address(reg, "ggml_backend_set_n_threads");
-                if (ggml_backend_set_n_threads_fn) {
-                    ctx->set_n_threads_fns.emplace_back(backend.get(), ggml_backend_set_n_threads_fn);
-                }
-            }
-        }
+    // this indicates we are doing pooled embedding, so we ignore batch.logits and output all tokens
+    const bool embd_pooled = cparams.embeddings && cparams.pooling_type != LLAMA_POOLING_TYPE_NONE;
 
-        llama_set_abort_callback(ctx, params.abort_callback, params.abort_callback_data);
+    lctx.embd_seq.clear();
 
-        if (!llama_kv_cache_init(ctx->kv_self, ctx, type_k, type_v, kv_size, cparams.offload_kqv)) {
-            LLAMA_LOG_ERROR("%s: llama_kv_cache_init() failed for self-attention cache\n", __func__);
-            llama_free(ctx);
-            return nullptr;
+    // count outputs
+    if (batch.logits && !embd_pooled) {
+        for (uint32_t i = 0; i < n_tokens_all; ++i) {
+            n_outputs += batch.logits[i] != 0;
         }
+    } else if (lctx.logits_all || embd_pooled) {
+        n_outputs = n_tokens_all;
+    } else {
+        // keep last output only
+        n_outputs = 1;
+    }
 
-        {
-            size_t memory_size_k = 0;
-            size_t memory_size_v = 0;
+    lctx.sbatch.from_batch(batch, n_embd,
+        /* simple_split */ !kv_self.recurrent,
+        /* logits_all   */ n_outputs == n_tokens_all);
 
-            for (auto & k : ctx->kv_self.k_l) {
-                memory_size_k += ggml_nbytes(k);
-            }
+    // reserve output buffer
+    if (llama_output_reserve(lctx, n_outputs) < n_outputs) {
+        LLAMA_LOG_ERROR("%s: could not reserve space for batch with %u outputs\n", __func__, n_outputs);
+        return -2;
+    };
 
-            for (auto & v : ctx->kv_self.v_l) {
-                memory_size_v += ggml_nbytes(v);
+    while (lctx.sbatch.n_tokens > 0) {
+        llama_ubatch ubatch;
+        if (kv_self.recurrent) {
+            if (embd_pooled) {
+                // Pooled embeddings cannot be split across ubatches (yet)
+                ubatch = lctx.sbatch.split_seq(n_ubatch);
+            } else {
+                // recurrent model architectures are easier to implement
+                // with equal-length sequences
+                ubatch = lctx.sbatch.split_equal(n_ubatch);
             }
-
-            LLAMA_LOG_INFO("%s: KV self size  = %7.2f MiB, K (%s): %7.2f MiB, V (%s): %7.2f MiB\n", __func__,
-                      (float)(memory_size_k + memory_size_v) / (1024.0f * 1024.0f),
-                ggml_type_name(type_k), (float)memory_size_k / (1024.0f * 1024.0f),
-                ggml_type_name(type_v), (float)memory_size_v / (1024.0f * 1024.0f));
+        } else {
+            ubatch = lctx.sbatch.split_simple(n_ubatch);
         }
+        const uint32_t n_tokens = ubatch.n_tokens;
 
-        // graph outputs buffer
+        // count the outputs in this u_batch
         {
-            // resized during inference when a batch uses more outputs
-            if (llama_output_reserve(*ctx, params.n_seq_max) < params.n_seq_max) {
-                LLAMA_LOG_ERROR("%s: failed to reserve initial output buffer\n", __func__);
-                llama_free(ctx);
-                return nullptr;
-            }
-
-            LLAMA_LOG_INFO("%s: %10s  output buffer size = %8.2f MiB\n", __func__,
-                    ggml_backend_buffer_name(ctx->buf_output.get()),
-                    ggml_backend_buffer_get_size(ctx->buf_output.get()) / 1024.0 / 1024.0);
-        }
+            int32_t n_outputs_new = 0;
 
-        // scheduler and compute buffers
-        {
-            // buffer types used for the compute buffer of each backend
-            std::vector backend_buft;
-            std::vector backend_ptrs;
-            for (auto & backend : ctx->backends) {
-                auto * buft = ggml_backend_get_default_buffer_type(backend.get());
-                auto backend_type = ggml_backend_dev_type(ggml_backend_get_device(backend.get()));
-                if (backend_type == GGML_BACKEND_DEVICE_TYPE_CPU && !model->devices.empty()) {
-                    // use the host buffer of the first device CPU for faster transfer of the intermediate state
-                    auto * dev = model->devices[0];
-                    auto * host_buft = ggml_backend_dev_host_buffer_type(dev);
-                    if (host_buft) {
-                        buft = host_buft;
-                    }
+            if (n_outputs == n_tokens_all) {
+                n_outputs_new = n_tokens;
+            } else {
+                GGML_ASSERT(ubatch.output);
+                for (uint32_t i = 0; i < n_tokens; i++) {
+                    n_outputs_new += (int32_t) (ubatch.output[i] != 0);
                 }
-                backend_buft.push_back(buft);
-                backend_ptrs.push_back(backend.get());
             }
 
-            const size_t max_nodes = llama_model_max_nodes(*model);
+            // needs to happen before the graph is built
+            lctx.n_outputs = n_outputs_new;
+        }
 
-            // buffer used to store the computation graph and the tensor meta data
-            ctx->buf_compute_meta.resize(ggml_tensor_overhead()*max_nodes + ggml_graph_overhead_custom(max_nodes, false));
+        int n_threads = n_tokens == 1 ? cparams.n_threads : cparams.n_threads_batch;
+        ggml_threadpool_t threadpool = n_tokens == 1 ? lctx.threadpool : lctx.threadpool_batch;
 
-            // TODO: move these checks to ggml_backend_sched
-            // enabling pipeline parallelism in the scheduler increases memory usage, so it is only done when necessary
-            bool pipeline_parallel =
-                llama_get_device_count(*model) > 1 &&
-                model->n_gpu_layers > (int)model->hparams.n_layer &&
-                model->split_mode == LLAMA_SPLIT_MODE_LAYER &&
-                params.offload_kqv;
+        GGML_ASSERT(n_threads > 0);
 
-            // pipeline parallelism requires support for async compute and events in all devices
-            if (pipeline_parallel) {
-                for (auto & backend : ctx->backends) {
-                    auto dev_type = ggml_backend_dev_type(ggml_backend_get_device(backend.get()));
-                    if (dev_type == GGML_BACKEND_DEVICE_TYPE_CPU) {
-                        // ignore CPU backend
-                        continue;
-                    }
-                    auto * dev = ggml_backend_get_device(backend.get());
-                    ggml_backend_dev_props props;
-                    ggml_backend_dev_get_props(dev, &props);
-                    if (!props.caps.async || !props.caps.events) {
-                        // device does not support async compute or events
-                        pipeline_parallel = false;
-                        break;
-                    }
-                }
-            }
+        // non-causal masks do not use the KV cache
+        if (hparams.causal_attn) {
+            llama_kv_cache_update(&lctx);
 
-            ctx->sched.reset(ggml_backend_sched_new(backend_ptrs.data(), backend_buft.data(), backend_ptrs.size(), max_nodes, pipeline_parallel));
+            // if we have enough unused cells before the current head ->
+            //   better to start searching from the beginning of the cache, hoping to fill it
+            if (kv_self.head > kv_self.used + 2*n_tokens) {
+                kv_self.head = 0;
+            }
 
-            if (pipeline_parallel) {
-                LLAMA_LOG_INFO("%s: pipeline parallelism enabled (n_copies=%d)\n", __func__, ggml_backend_sched_get_n_copies(ctx->sched.get()));
+            const auto slot = llama_kv_cache_find_slot(kv_self, ubatch);
+            if (!slot) {
+                return 1;
             }
+            kv_slot_restorer.save(slot);
 
-            // initialize scheduler with the worst-case graph
-            uint32_t n_seqs = 1; // TODO: worst-case number of sequences
-            uint32_t n_tokens = std::min(cparams.n_ctx, cparams.n_ubatch);
-            llama_token token = llama_token_bos(&ctx->model); // not actually used by llama_build_graph, but required to choose between token and embedding inputs graph
+            if (!kv_self.recurrent) {
+                // a heuristic, to avoid attending the full cache if it is not yet utilized
+                // after enough generations, the benefit from this heuristic disappears
+                // if we start defragmenting the cache, the benefit from this will be more important
+                const uint32_t pad = llama_kv_cache_get_padding(cparams);
+                kv_self.n = std::min(kv_self.size, std::max(pad, GGML_PAD(llama_kv_cache_cell_max(kv_self), pad)));
+                //kv_self.n = llama_kv_cache_cell_max(kv_self);
+            }
+        }
 
-            llama_ubatch ubatch_pp = { true, n_tokens, n_tokens / n_seqs, n_seqs, &token, nullptr, nullptr, nullptr, nullptr, nullptr};
-            ggml_cgraph * gf_pp = llama_build_graph(*ctx, ubatch_pp, true);
+        //printf("kv_self.n = %5d, kv_self.used = %5d, kv_self.head = %5d\n", kv_self.n, kv_self.used, kv_self.head);
 
-            // reserve pp graph first so that buffers are only allocated once
-            ggml_backend_sched_reserve(ctx->sched.get(), gf_pp);
-            int n_splits_pp = ggml_backend_sched_get_n_splits(ctx->sched.get());
-            int n_nodes_pp = ggml_graph_n_nodes(gf_pp);
+        ggml_backend_sched_reset(lctx.sched.get());
+        ggml_backend_sched_set_eval_callback(lctx.sched.get(), lctx.cparams.cb_eval, lctx.cparams.cb_eval_user_data);
 
-            // reserve with tg graph to get the number of splits and nodes
-            llama_ubatch ubatch_tg = { true, 1, 1, n_seqs, &token, nullptr, nullptr, nullptr, nullptr, nullptr};
-            ggml_cgraph * gf_tg = llama_build_graph(*ctx, ubatch_tg, true);
-            ggml_backend_sched_reserve(ctx->sched.get(), gf_tg);
-            int n_splits_tg = ggml_backend_sched_get_n_splits(ctx->sched.get());
-            int n_nodes_tg = ggml_graph_n_nodes(gf_tg);
+        ggml_cgraph * gf = llama_build_graph(lctx, ubatch, false);
 
-            // reserve again with pp graph to avoid ggml-alloc reallocations during inference
-            gf_pp = llama_build_graph(*ctx, ubatch_pp, true);
-            if (!ggml_backend_sched_reserve(ctx->sched.get(), gf_pp)) {
-                LLAMA_LOG_ERROR("%s: failed to allocate compute buffers\n", __func__);
-                llama_free(ctx);
-                return nullptr;
-            }
+        // the output is always the last tensor in the graph
+        struct ggml_tensor * res  = ggml_graph_node(gf, -1);
+        struct ggml_tensor * embd = ggml_graph_node(gf, -2);
 
-            for (size_t i = 0; i < backend_ptrs.size(); ++i) {
-                ggml_backend_t backend = backend_ptrs[i];
-                ggml_backend_buffer_type_t buft = backend_buft[i];
-                size_t size = ggml_backend_sched_get_buffer_size(ctx->sched.get(), backend);
-                if (size > 1) {
-                    LLAMA_LOG_INFO("%s: %10s compute buffer size = %8.2f MiB\n", __func__,
-                            ggml_backend_buft_name(buft),
-                            size / 1024.0 / 1024.0);
+        if (lctx.n_outputs == 0) {
+            // no output
+            res  = nullptr;
+            embd = nullptr;
+        } else if (cparams.embeddings) {
+            res  = nullptr; // do not extract logits for embedding case
+            embd = nullptr;
+            for (int i = ggml_graph_n_nodes(gf) - 1; i >= 0; --i) {
+                if (strcmp(ggml_graph_node(gf, i)->name, "result_embd_pooled") == 0) {
+                    embd = ggml_graph_node(gf, i);
+                    break;
                 }
             }
-
-            if (n_nodes_pp == n_nodes_tg) {
-                LLAMA_LOG_INFO("%s: graph nodes  = %d\n", __func__, n_nodes_pp);
-            } else {
-                LLAMA_LOG_INFO("%s: graph nodes  = %d (with bs=%d), %d (with bs=1)\n", __func__, n_nodes_pp, n_tokens, n_nodes_tg);
-            }
-            if (n_splits_pp == n_splits_tg) {
-                LLAMA_LOG_INFO("%s: graph splits = %d\n", __func__, n_splits_pp);
-            } else {
-                LLAMA_LOG_INFO("%s: graph splits = %d (with bs=%d), %d (with bs=1)\n", __func__, n_splits_pp, n_tokens, n_splits_tg);
-            }
+            GGML_ASSERT(embd != nullptr && "missing embeddings tensor");
+        } else {
+            embd = nullptr; // do not extract embeddings when not needed
+            GGML_ASSERT(strcmp(res->name, "result_output") == 0 && "missing result_output tensor");
         }
-    }
 
-    return ctx;
-}
+        // LLAMA_LOG_INFO("graph build time: %.3f ms (%d nodes, %d leafs)\n", (ggml_time_us() - t_start_us)/1000.0, gf->n_nodes, gf->n_leafs);
 
-void llama_free(struct llama_context * ctx) {
-    delete ctx;
-}
+        ggml_backend_sched_alloc_graph(lctx.sched.get(), gf);
 
-uint32_t llama_n_ctx(const struct llama_context * ctx) {
-    return ctx->cparams.n_ctx;
-}
+        llama_set_inputs(lctx, ubatch);
 
-uint32_t llama_n_batch(const struct llama_context * ctx) {
-    return ctx->cparams.n_batch;
-}
+        const auto compute_status = llama_graph_compute(lctx, gf, n_threads, threadpool);
+        if (compute_status != GGML_STATUS_SUCCESS) {
+            kv_slot_restorer.restore(kv_self);
+            switch (compute_status) {
+                case GGML_STATUS_ABORTED:
+                    return 2;
+                case GGML_STATUS_ALLOC_FAILED:
+                    return -2;
+                case GGML_STATUS_FAILED:
+                default:
+                    return -3;
+            }
+        }
 
-uint32_t llama_n_ubatch(const struct llama_context * ctx) {
-    return ctx->cparams.n_ubatch;
-}
+        // update the kv ring buffer
+        {
+            kv_self.head += n_tokens;
 
-uint32_t llama_n_seq_max(const struct llama_context * ctx) {
-    return ctx->kv_self.size;
-}
+            // Ensure kv cache head points to a valid index.
+            if (kv_self.head >= kv_self.size) {
+                kv_self.head = 0;
+            }
+        }
 
-enum llama_vocab_type llama_vocab_type(const struct llama_model * model) {
-    return model->vocab.type;
-}
+        // plot the computation graph in dot format (for debugging purposes)
+        //if (n_past%100 == 0) {
+        //    ggml_graph_dump_dot(gf, NULL, "llama.dot");
+        //}
 
-int32_t llama_n_vocab(const struct llama_model * model) {
-    return model->hparams.n_vocab;
-}
+        // extract logits
+        if (res) {
+            ggml_backend_t backend_res = ggml_backend_sched_get_tensor_backend(lctx.sched.get(), res);
+            GGML_ASSERT(backend_res != nullptr);
+            GGML_ASSERT(lctx.logits != nullptr);
 
-int32_t llama_n_ctx_train(const struct llama_model * model) {
-    return model->hparams.n_ctx_train;
-}
+            float * logits_out = lctx.logits + n_outputs_prev*n_vocab;
+            const int32_t n_outputs_new = lctx.n_outputs;
 
-int32_t llama_n_embd(const struct llama_model * model) {
-    return model->hparams.n_embd;
-}
+            if (n_outputs_new) {
+                GGML_ASSERT( n_outputs_prev + n_outputs_new <= n_outputs);
+                GGML_ASSERT((n_outputs_prev + n_outputs_new)*n_vocab <= (int64_t) lctx.logits_size);
+                ggml_backend_tensor_get_async(backend_res, res, logits_out, 0, n_outputs_new*n_vocab*sizeof(float));
+            }
+        }
 
-int32_t llama_n_layer(const struct llama_model * model) {
-    return model->hparams.n_layer;
-}
+        // extract embeddings
+        if (embd) {
+            ggml_backend_t backend_embd = ggml_backend_sched_get_tensor_backend(lctx.sched.get(), embd);
+            GGML_ASSERT(backend_embd != nullptr);
 
-int32_t llama_n_head(const struct llama_model * model) {
-    return model->hparams.n_head();
-}
+            switch (cparams.pooling_type) {
+                case LLAMA_POOLING_TYPE_NONE:
+                    {
+                        // extract token embeddings
+                        GGML_ASSERT(lctx.embd != nullptr);
+                        float * embd_out = lctx.embd + n_outputs_prev*n_embd;
+                        const int32_t n_outputs_new = lctx.n_outputs;
 
-const struct llama_model * llama_get_model(const struct llama_context * ctx) {
-    return &ctx->model;
-}
+                        if (n_outputs_new) {
+                            GGML_ASSERT( n_outputs_prev + n_outputs_new <= n_outputs);
+                            GGML_ASSERT((n_outputs_prev + n_outputs_new)*n_embd <= (int64_t) lctx.embd_size);
+                            ggml_backend_tensor_get_async(backend_embd, embd, embd_out, 0, n_outputs_new*n_embd*sizeof(float));
+                        }
+                    } break;
+                case LLAMA_POOLING_TYPE_MEAN:
+                case LLAMA_POOLING_TYPE_CLS:
+                case LLAMA_POOLING_TYPE_LAST:
+                    {
+                        // extract sequence embeddings (cleared before processing each batch)
+                        auto & embd_seq_out = lctx.embd_seq;
 
-enum llama_pooling_type llama_pooling_type(const struct llama_context * ctx) {
-    return ctx->cparams.pooling_type;
-}
+                        for (uint32_t s = 0; s < ubatch.n_seqs; ++s) {
+                            const llama_seq_id seq_id = ubatch.seq_id[s][0];
+                            if (embd_seq_out.find(seq_id) != embd_seq_out.end()) {
+                                continue;
+                            }
+                            embd_seq_out[seq_id].resize(n_embd);
+                            ggml_backend_tensor_get_async(backend_embd, embd, embd_seq_out[seq_id].data(), (n_embd*seq_id)*sizeof(float), n_embd*sizeof(float));
+                        }
+                    } break;
+                case LLAMA_POOLING_TYPE_RANK:
+                    {
+                        // extract the rerank score - a single float per sequence
+                        auto & embd_seq_out = lctx.embd_seq;
 
-enum llama_rope_type llama_rope_type(const struct llama_model * model) {
-    switch (model->arch) {
-        // these models do not use RoPE
-        case LLM_ARCH_GPT2:
-        case LLM_ARCH_GPTJ:
-        case LLM_ARCH_MPT:
-        case LLM_ARCH_REFACT:
-        case LLM_ARCH_BLOOM:
-        case LLM_ARCH_MAMBA:
-        case LLM_ARCH_JINA_BERT_V2:
-        case LLM_ARCH_T5:
-        case LLM_ARCH_T5ENCODER:
-        case LLM_ARCH_JAIS:
-        case LLM_ARCH_RWKV6:
-            return LLAMA_ROPE_TYPE_NONE;
+                        for (uint32_t s = 0; s < ubatch.n_seqs; ++s) {
+                            const llama_seq_id seq_id = ubatch.seq_id[s][0];
+                            if (embd_seq_out.find(seq_id) != embd_seq_out.end()) {
+                                continue;
+                            }
+                            embd_seq_out[seq_id].resize(1);
+                            ggml_backend_tensor_get_async(backend_embd, embd, embd_seq_out[seq_id].data(), (seq_id)*sizeof(float), sizeof(float));
+                        }
+                    } break;
+                case LLAMA_POOLING_TYPE_UNSPECIFIED:
+                    {
+                        GGML_ABORT("unknown pooling type");
+                    }
+            }
+        }
+        n_outputs_prev += lctx.n_outputs;
+    }
 
-        // use what we call a normal RoPE, operating on pairs of consecutive head values
-        case LLM_ARCH_LLAMA:
-        case LLM_ARCH_BAICHUAN:
-        case LLM_ARCH_STARCODER:
-        case LLM_ARCH_PLAMO:
-        case LLM_ARCH_ORION:
-        case LLM_ARCH_INTERNLM2:
-        case LLM_ARCH_MINICPM:
-        case LLM_ARCH_XVERSE:
-        case LLM_ARCH_COMMAND_R:
-        case LLM_ARCH_OLMO:
-        case LLM_ARCH_ARCTIC:
-        case LLM_ARCH_DEEPSEEK:
-        case LLM_ARCH_DEEPSEEK2:
-        case LLM_ARCH_CHATGLM:
-        case LLM_ARCH_GRANITE:
-        case LLM_ARCH_GRANITE_MOE:
-        case LLM_ARCH_CHAMELEON:
-            return LLAMA_ROPE_TYPE_NORM;
+    // set output mappings
+    {
+        bool sorted_output = true;
 
-        // the pairs of head values are offset by n_rot/2
-        case LLM_ARCH_FALCON:
-        case LLM_ARCH_GROK:
-        case LLM_ARCH_DBRX:
-        case LLM_ARCH_BERT:
-        case LLM_ARCH_NOMIC_BERT:
-        case LLM_ARCH_STABLELM:
-        case LLM_ARCH_BITNET:
-        case LLM_ARCH_QWEN:
-        case LLM_ARCH_QWEN2:
-        case LLM_ARCH_QWEN2MOE:
-        case LLM_ARCH_OLMO2:
-        case LLM_ARCH_OLMOE:
-        case LLM_ARCH_PHI2:
-        case LLM_ARCH_PHI3:
-        case LLM_ARCH_GEMMA:
-        case LLM_ARCH_GEMMA2:
-        case LLM_ARCH_STARCODER2:
-        case LLM_ARCH_OPENELM:
-        case LLM_ARCH_GPTNEOX:
-        case LLM_ARCH_CODESHELL:
-        case LLM_ARCH_NEMOTRON:
-        case LLM_ARCH_EXAONE:
-        case LLM_ARCH_MINICPM3:
-            return LLAMA_ROPE_TYPE_NEOX;
+        GGML_ASSERT(lctx.sbatch.out_ids.size() == n_outputs);
 
-        case LLM_ARCH_QWEN2VL:
-            return LLAMA_ROPE_TYPE_MROPE;
+        for (size_t i = 0; i < n_outputs; ++i) {
+            size_t out_id = lctx.sbatch.out_ids[i];
+            lctx.output_ids[out_id] = i;
+            if (out_id != i) {
+                sorted_output = false;
+            }
+        }
 
-        // all model arches should be listed explicitly here
-        case LLM_ARCH_UNKNOWN:
-            GGML_ABORT("unknown architecture");
+        if (sorted_output) {
+            lctx.sbatch.out_ids.clear();
+        }
     }
 
-    return LLAMA_ROPE_TYPE_NONE;
-}
+    // set to total number of outputs in the batch, for use in llama_get_logits_ith
+    lctx.n_outputs = n_outputs;
 
-float llama_rope_freq_scale_train(const struct llama_model * model) {
-    return model->hparams.rope_freq_scale_train;
-}
+    // wait for the computation to finish (automatically done when obtaining the model output)
+    //llama_synchronize(&lctx);
 
-int32_t llama_model_meta_val_str(const struct llama_model * model, const char * key, char * buf, size_t buf_size) {
-    const auto & it = model->gguf_kv.find(key);
-    if (it == model->gguf_kv.end()) {
-        if (buf_size > 0) {
-            buf[0] = '\0';
-        }
-        return -1;
-    }
-    return snprintf(buf, buf_size, "%s", it->second.c_str());
-}
+    // decide if we need to defrag the kv cache
+    if (cparams.causal_attn && cparams.defrag_thold >= 0.0f) {
+        const float fragmentation = kv_self.n >= 128 ? 1.0f - float(kv_self.used)/float(kv_self.n) : 0.0f;
 
-int32_t llama_model_meta_count(const struct llama_model * model) {
-    return (int)model->gguf_kv.size();
-}
+        // queue defragmentation for next llama_kv_cache_update
+        if (fragmentation > cparams.defrag_thold) {
+            //LLAMA_LOG_INFO("fragmentation: %.2f\n", fragmentation);
 
-int32_t llama_model_meta_key_by_index(const struct llama_model * model, int i, char * buf, size_t buf_size) {
-    if (i < 0 || i >= (int)model->gguf_kv.size()) {
-        if (buf_size > 0) {
-            buf[0] = '\0';
+            llama_kv_cache_defrag(kv_self);
         }
-        return -1;
     }
-    auto it = model->gguf_kv.begin();
-    std::advance(it, i);
-    return snprintf(buf, buf_size, "%s", it->first.c_str());
-}
 
-int32_t llama_model_meta_val_str_by_index(const struct llama_model * model, int32_t i, char * buf, size_t buf_size) {
-    if (i < 0 || i >= (int)model->gguf_kv.size()) {
-        if (buf_size > 0) {
-            buf[0] = '\0';
-        }
-        return -1;
-    }
-    auto it = model->gguf_kv.begin();
-    std::advance(it, i);
-    return snprintf(buf, buf_size, "%s", it->second.c_str());
-}
+    // Reset state for the next token before backend sync, to allow the CPU activities in the reset to
+    // overlap with device computation.
+    ggml_backend_sched_reset(lctx.sched.get());
 
-int32_t llama_model_desc(const struct llama_model * model, char * buf, size_t buf_size) {
-    return snprintf(buf, buf_size, "%s %s %s",
-            llama_model_arch_name(model->arch),
-            llama_model_type_name(model->type),
-            llama_model_ftype_name(model->ftype).c_str());
+    return 0;
 }
 
-uint64_t llama_model_size(const struct llama_model * model) {
-    return model->n_bytes;
-}
+// encode a batch of tokens by evaluating the encoder part of the transformer
+//
+//   - lctx:      llama context
+//   - batch:     batch to evaluate
+//
+// return 0 on success
+// return positive int on warning
+// return negative int on error
+//
+static int llama_encode_impl(
+         llama_context & lctx,
+           llama_batch   inp_batch) {
 
-uint64_t llama_model_n_params(const struct llama_model * model) {
-    return model->n_elements;
-}
+    lctx.is_encoding = true;
 
-struct ggml_tensor * llama_get_model_tensor(struct llama_model * model, const char * name) {
-    auto it = std::find_if(model->tensors_by_name.begin(), model->tensors_by_name.end(),
-            [name](const std::pair & it) {
-                return it.first == name;
-            });
-    if (it == model->tensors_by_name.end()) {
-        return nullptr;
+    if (inp_batch.n_tokens == 0) {
+        LLAMA_LOG_ERROR("%s: n_tokens == 0\n", __func__);
+        return -1;
     }
-    return it->second;
-}
 
-bool llama_model_has_encoder(const struct llama_model * model) {
-    switch (model->arch) {
-        case LLM_ARCH_T5:        return true;
-        case LLM_ARCH_T5ENCODER: return true;
-        default:                 return false;
-    }
-}
+    // temporary allocate memory for the input batch if needed
+    llama_batch_allocr batch_allocr(inp_batch, inp_batch.pos ? -1 : lctx.kv_self.max_pos() + 1);
+
+    const llama_batch & batch = batch_allocr.batch;
+    const uint32_t n_tokens = batch.n_tokens;
+
+    const auto & model   = lctx.model;
+    const auto & hparams = model.hparams;
+    const auto & cparams = lctx.cparams;
+
+    GGML_ASSERT((!batch.token && batch.embd) || (batch.token && !batch.embd)); // NOLINT
 
-bool llama_model_has_decoder(const struct llama_model * model) {
-    switch (model->arch) {
-        case LLM_ARCH_T5ENCODER: return false;
-        default:                 return true;
+    if (batch.token) {
+        for (uint32_t i = 0; i < n_tokens; ++i) {
+            if (batch.token[i] < 0 || (uint32_t)batch.token[i] >= model.vocab.n_vocab) {
+                LLAMA_LOG_ERROR("%s: invalid token[%d] = %d\n", __func__, i, batch.token[i]);
+                return -1;
+            }
+        }
     }
-}
 
-llama_token llama_model_decoder_start_token(const struct llama_model * model) {
-    return model->hparams.dec_start_token_id;
-}
+    // micro-batching is not possible for non-causal encoding, so we process the batch in a single shot
+    GGML_ASSERT(cparams.n_ubatch >= n_tokens && "encoder requires n_ubatch >= n_tokens");
 
-bool llama_model_is_recurrent(const struct llama_model * model) {
-    switch (model->arch) {
-        case LLM_ARCH_MAMBA:  return true;
-        case LLM_ARCH_RWKV6:  return true;
-        default:              return false;
+    if (lctx.t_compute_start_us == 0) {
+        lctx.t_compute_start_us = ggml_time_us();
     }
-}
 
-uint32_t llama_model_quantize(
-        const char * fname_inp,
-        const char * fname_out,
-        const llama_model_quantize_params * params) {
-    try {
-        llama_model_quantize_internal(fname_inp, fname_out, params);
-        return 0;
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: failed to quantize: %s\n", __func__, err.what());
-        return 1;
-    }
-}
+    lctx.n_queued_tokens += n_tokens;
 
-struct llama_lora_adapter * llama_lora_adapter_init(struct llama_model * model, const char * path_lora) {
-    try {
-        struct llama_lora_adapter * adapter = new llama_lora_adapter(model);
-        llama_lora_adapter_init_internal(model, path_lora, *adapter);
-        return adapter;
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: failed to apply lora adapter: %s\n", __func__, err.what());
-        return nullptr;
-    }
-}
+    const int64_t n_embd = hparams.n_embd;
 
-static bool llama_control_vector_init(struct llama_control_vector & cvec, const llama_model & model) {
-    GGML_ASSERT(cvec.tensors.empty());
-    GGML_ASSERT(cvec.ctxs.empty());
-    GGML_ASSERT(cvec.bufs.empty());
+    lctx.sbatch.from_batch(batch, n_embd, /* simple_split */ true, /* logits_all */ true);
 
-    // create a context for each buffer type
-    std::map ctx_map;
-    auto ctx_for_buft = [&](ggml_backend_buffer_type_t buft) -> ggml_context * {
-        auto it = ctx_map.find(buft);
-        if (it == ctx_map.end()) {
-            struct ggml_init_params params = {
-                /*.mem_size   =*/ model.hparams.n_layer*ggml_tensor_overhead(),
-                /*.mem_buffer =*/ NULL,
-                /*.no_alloc   =*/ true,
-            };
-            ggml_context * ctx = ggml_init(params);
-            if (!ctx) {
-                return nullptr;
-            }
-            ctx_map[buft] = ctx;
-            cvec.ctxs.emplace_back(ctx);
-            return ctx;
-        }
-        return it->second;
+    const llama_ubatch ubatch = lctx.sbatch.split_simple(n_tokens);
+
+    // reserve output buffer
+    if (llama_output_reserve(lctx, n_tokens) < n_tokens) {
+        LLAMA_LOG_ERROR("%s: could not reserve space for batch with %u outputs\n", __func__, n_tokens);
+        return -2;
     };
 
-    // make tensors
-    cvec.tensors.reserve(model.hparams.n_layer);
-    cvec.tensors.push_back(nullptr); // there's never a tensor for layer 0
-    for (size_t il = 1; il < model.hparams.n_layer; il++) {
-        ggml_backend_buffer_type_t buft = select_buft(*model.dev_layer.at(il).buft_list,
-            [&](ggml_context * ctx) {
-                ggml_tensor * cur = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, model.hparams.n_embd);
-                ggml_tensor * layer_dir = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, model.hparams.n_embd);
-                return ggml_add(ctx, cur, layer_dir);
-            });
-        ggml_context * ctx = ctx_for_buft(buft);
-        if (!ctx) {
-            LLAMA_LOG_ERROR("%s: failed to allocate context for control vector\n", __func__);
-            return false;
-        }
-        ggml_tensor * tensor = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, model.hparams.n_embd);
-        cvec.tensors.push_back(tensor);
+    for (uint32_t i = 0; i < n_tokens; ++i) {
+        lctx.output_ids[i] = i;
     }
 
-    // allocate tensors / buffers and zero
-    cvec.bufs.reserve(ctx_map.size());
-    for (auto it : ctx_map) {
-        ggml_backend_buffer_type_t buft = it.first;
-        ggml_context * ctx = it.second;
-        ggml_backend_buffer_t buf = ggml_backend_alloc_ctx_tensors_from_buft(ctx, buft);
-        if (!buf) {
-            LLAMA_LOG_ERROR("%s: failed to allocate buffer for control vector\n", __func__);
-            return false;
-        }
-        ggml_backend_buffer_clear(buf, 0);
-        cvec.bufs.emplace_back(buf);
-    }
+    lctx.inp_embd_enc = NULL;
+    lctx.n_outputs = n_tokens;
 
-    return true;
-}
+    int n_threads = n_tokens == 1 ? cparams.n_threads : cparams.n_threads_batch;
+    ggml_threadpool_t threadpool = n_tokens == 1 ? lctx.threadpool : lctx.threadpool_batch;
+
+    GGML_ASSERT(n_threads > 0);
 
-int32_t llama_control_vector_apply(struct llama_context * lctx, const float * data, size_t len, int32_t n_embd, int32_t il_start, int32_t il_end) {
-    const llama_model & model = lctx->model;
-    llama_control_vector & cvec = lctx->cvec;
+    ggml_backend_sched_reset(lctx.sched.get());
+    ggml_backend_sched_set_eval_callback(lctx.sched.get(), lctx.cparams.cb_eval, lctx.cparams.cb_eval_user_data);
 
-    if (data == nullptr) {
-        // disable the current control vector (but leave allocated for later)
-        cvec.layer_start = -1;
-        cvec.layer_end   = -1;
-        return 0;
-    }
+    ggml_cgraph * gf = llama_build_graph(lctx, ubatch, false);
 
-    if (n_embd != (int) model.hparams.n_embd) {
-        LLAMA_LOG_ERROR("%s: control vector n_embd does not match model\n", __func__);
-        return 1;
-    }
+    // the output embeddings after the final encoder normalization
+    struct ggml_tensor * embd = nullptr;
 
-    if (cvec.tensors.empty()) {
-        if (!llama_control_vector_init(cvec, model)) {
-            return 1;
+    // there are two cases here
+    if (llama_model_has_decoder(&lctx.model)) {
+        // first case is an encoder-decoder T5 model where embeddings are passed to decoder
+        embd = ggml_graph_node(gf, -1);
+        GGML_ASSERT(strcmp(embd->name, "result_norm") == 0 && "missing result_output tensor");
+    } else {
+        // second case is an encoder-only T5 model
+        if (cparams.embeddings) {
+            // only output embeddings if required
+            embd = ggml_graph_node(gf, -1);
+            if (strcmp(embd->name, "result_embd_pooled") != 0) {
+                embd = ggml_graph_node(gf, -2);
+            }
+            GGML_ASSERT(strcmp(embd->name, "result_embd_pooled") == 0 && "missing embeddings tensor");
         }
     }
 
-    cvec.layer_start = il_start;
-    cvec.layer_end   = il_end;
+    ggml_backend_sched_alloc_graph(lctx.sched.get(), gf);
 
-    for (size_t il = 1; il < model.hparams.n_layer; il++) {
-        assert(cvec.tensors[il] != nullptr);
+    llama_set_inputs(lctx, ubatch);
 
-        const size_t off = n_embd * (il - 1); // buffer doesn't have data for layer 0, since it's never present
-        if (off + n_embd <= len) {
-            ggml_backend_tensor_set(cvec.tensors[il], data + off, 0, n_embd * ggml_element_size(cvec.tensors[il]));
-        }
+    const auto compute_status = llama_graph_compute(lctx, gf, n_threads, threadpool);
+    switch (compute_status) {
+        case GGML_STATUS_SUCCESS:
+            break;
+        case GGML_STATUS_ABORTED:
+            return 2;
+        case GGML_STATUS_ALLOC_FAILED:
+            return -2;
+        case GGML_STATUS_FAILED:
+        default:
+            return -3;
     }
 
-    return 0;
-}
-
-struct llama_kv_cache_view llama_kv_cache_view_init(const struct llama_context * ctx, int32_t n_seq_max) {
-    struct llama_kv_cache_view result = {
-        /*.n_cells            = */ 0,
-        /*.n_seq_max          = */ n_seq_max,
-        /*.token_count        = */ 0,
-        /*.used_cells         = */ llama_get_kv_cache_used_cells(ctx),
-        /*.max_contiguous     = */ 0,
-        /*.max_contiguous_idx = */ -1,
-        /*.cells              = */ nullptr,
-        /*.cells_sequences    = */ nullptr,
-    };
-    return result;
-}
+    // extract embeddings
+    if (embd) {
+        ggml_backend_t backend_embd = ggml_backend_sched_get_tensor_backend(lctx.sched.get(), embd);
+        GGML_ASSERT(backend_embd != nullptr);
 
-void llama_kv_cache_view_free(struct llama_kv_cache_view * view) {
-    if (view->cells != nullptr) {
-        free(view->cells);
-        view->cells = nullptr;
-    }
-    if (view->cells_sequences != nullptr) {
-        free(view->cells_sequences);
-        view->cells_sequences = nullptr;
-    }
-}
+        if (llama_model_has_decoder(&lctx.model)) {
+            lctx.embd_enc.resize(n_tokens*n_embd);
+            float * embd_out = lctx.embd_enc.data();
 
-void llama_kv_cache_view_update(const struct llama_context * ctx, struct llama_kv_cache_view * view) {
-    if (uint32_t(view->n_cells) < ctx->kv_self.size || view->cells == nullptr) {
-        view->n_cells = int32_t(ctx->kv_self.size);
-        void * p = realloc(view->cells, sizeof(struct llama_kv_cache_view_cell) * view->n_cells);
-        GGML_ASSERT(p != nullptr && "Failed to alloc kv_cache_view cells");
-        view->cells = (struct llama_kv_cache_view_cell *)p;
-        p = realloc(view->cells_sequences, sizeof(llama_seq_id) * view->n_seq_max * view->n_cells);
-        GGML_ASSERT(p != nullptr && "Failed to alloc kv_cache_view cells sequences");
-        view->cells_sequences = (llama_seq_id *)p;
-    }
+            ggml_backend_tensor_get_async(backend_embd, embd, embd_out, 0, n_tokens*n_embd*sizeof(float));
+            GGML_ASSERT(!ubatch.equal_seqs); // TODO: handle equal splits
 
-    const std::vector & kv_cells = ctx->kv_self.cells;
-    llama_kv_cache_view_cell * c_curr = view->cells;
-    llama_seq_id * cs_curr = view->cells_sequences;
-    int32_t used_cells = 0;
-    int32_t token_count = 0;
-    int32_t curr_contig_idx = -1;
-    uint32_t max_contig = 0;
-    int32_t max_contig_idx = -1;
-
-    for (int32_t i = 0; i < int32_t(ctx->kv_self.size); i++, c_curr++, cs_curr += view->n_seq_max) {
-        const size_t curr_size = kv_cells[i].seq_id.size();
-        token_count += curr_size;
-        c_curr->pos = kv_cells[i].pos + kv_cells[i].delta;
-
-        if (curr_size > 0) {
-            if (curr_contig_idx >= 0 && uint32_t(i - curr_contig_idx) > max_contig) {
-                max_contig = i - curr_contig_idx;
-                max_contig_idx = curr_contig_idx;
+            // remember the sequence ids used during the encoding - needed for cross attention later
+            lctx.seq_ids_enc.resize(n_tokens);
+            for (uint32_t i = 0; i < n_tokens; i++) {
+                for (int s = 0; s < ubatch.n_seq_id[i]; s++) {
+                    llama_seq_id seq_id = ubatch.seq_id[i][s];
+                    lctx.seq_ids_enc[i].insert(seq_id);
+                }
             }
-            curr_contig_idx = -1;
-        } else if (curr_contig_idx < 0) {
-            curr_contig_idx = i;
-        }
+        } else {
+            GGML_ASSERT(lctx.embd != nullptr);
 
-        int seq_idx = 0;
-        for (const llama_seq_id it : kv_cells[i].seq_id) {
-            if (seq_idx >= view->n_seq_max) {
-                break;
+            switch (cparams.pooling_type) {
+                case LLAMA_POOLING_TYPE_NONE:
+                    {
+                        // extract token embeddings
+                        GGML_ASSERT(lctx.embd != nullptr);
+                        float * embd_out = lctx.embd;
+
+                        GGML_ASSERT(n_tokens*n_embd <= (int64_t) lctx.embd_size);
+                        ggml_backend_tensor_get_async(backend_embd, embd, embd_out, 0, n_tokens*n_embd*sizeof(float));
+                    } break;
+                case LLAMA_POOLING_TYPE_MEAN:
+                case LLAMA_POOLING_TYPE_CLS:
+                case LLAMA_POOLING_TYPE_LAST:
+                    {
+                        // extract sequence embeddings
+                        auto & embd_seq_out = lctx.embd_seq;
+                        embd_seq_out.clear();
+
+                        GGML_ASSERT(!ubatch.equal_seqs); // TODO: handle equal splits
+
+                        for (uint32_t i = 0; i < n_tokens; i++) {
+                            const llama_seq_id seq_id = ubatch.seq_id[i][0];
+                            if (embd_seq_out.find(seq_id) != embd_seq_out.end()) {
+                                continue;
+                            }
+                            embd_seq_out[seq_id].resize(n_embd);
+                            ggml_backend_tensor_get_async(backend_embd, embd, embd_seq_out[seq_id].data(), (n_embd*seq_id)*sizeof(float), n_embd*sizeof(float));
+                        }
+                    } break;
+                case LLAMA_POOLING_TYPE_RANK:
+                    {
+                        // TODO: this likely should be the same logic as in llama_decoder_internal, but better to
+                        //       wait for an encoder model that requires this pooling type in order to test it
+                        //       https://github.com/ggerganov/llama.cpp/pull/9510
+                        GGML_ABORT("RANK pooling not implemented yet");
+                    }
+                case LLAMA_POOLING_TYPE_UNSPECIFIED:
+                    {
+                        GGML_ABORT("unknown pooling type");
+                    }
             }
-            cs_curr[seq_idx] = it;
-            seq_idx++;
-        }
-        if (seq_idx != 0) {
-            used_cells++;
         }
-        for (; seq_idx < view->n_seq_max; seq_idx++) {
-            cs_curr[seq_idx] = -1;
-        }
-    }
-    if (curr_contig_idx >= 0 && kv_cells.size() - curr_contig_idx > max_contig) {
-        max_contig_idx = curr_contig_idx;
-        max_contig = kv_cells.size() - curr_contig_idx;
     }
-    view->max_contiguous = max_contig;
-    view->max_contiguous_idx = max_contig_idx;
-    view->token_count = token_count;
-    view->used_cells = used_cells;
-    if (uint32_t(used_cells) != ctx->kv_self.used) {
-        LLAMA_LOG_ERROR("%s: used cells mismatch. kv_cache says %d but we calculated %d\n",
-            __func__, ctx->kv_self.used, used_cells);
-    }
-}
-
-int32_t llama_get_kv_cache_token_count(const struct llama_context * ctx) {
-    int result = 0;
 
-    for (uint32_t i = 0; i < ctx->kv_self.size; i++) {
-        result += ctx->kv_self.cells[i].seq_id.size();
-    }
+    // Reset state for the next token before backend sync, to allow the CPU activities in the reset to
+    // overlap with device computation.
+    ggml_backend_sched_reset(lctx.sched.get());
 
-    return result;
+    return 0;
 }
 
-int32_t llama_get_kv_cache_used_cells(const struct llama_context * ctx) {
-    return ctx->kv_self.used;
-}
+// find holes from the beginning of the KV cache and fill them by moving data from the end of the cache
+static void llama_kv_cache_defrag_impl(struct llama_context & lctx) {
+    auto & kv_self = lctx.kv_self;
 
-void llama_kv_cache_clear(struct llama_context * ctx) {
-    llama_kv_cache_clear(ctx->kv_self);
-}
+    const auto & hparams = lctx.model.hparams;
 
-bool llama_kv_cache_seq_rm(struct llama_context * ctx, llama_seq_id seq_id, llama_pos p0, llama_pos p1) {
-    return llama_kv_cache_seq_rm(ctx->kv_self, seq_id, p0, p1);
-}
+    const uint32_t n_layer = hparams.n_layer;
 
-void llama_kv_cache_seq_cp(struct llama_context * ctx, llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) {
-    if (seq_id_src == seq_id_dst) {
-        return;
-    }
-    llama_kv_cache_seq_cp(ctx->kv_self, seq_id_src, seq_id_dst, p0, p1);
-}
+    const uint32_t n_kv   = llama_kv_cache_cell_max(kv_self);
+    const uint32_t n_used = kv_self.used;
 
-void llama_kv_cache_seq_keep(struct llama_context * ctx, llama_seq_id seq_id) {
-    llama_kv_cache_seq_keep(ctx->kv_self, seq_id);
-}
+    assert(n_used <= n_kv);
 
-void llama_kv_cache_seq_add(struct llama_context * ctx, llama_seq_id seq_id, llama_pos p0, llama_pos p1, llama_pos delta) {
-    if (delta == 0) {
-        return;
-    }
+    //const int64_t t_start = ggml_time_us();
 
-    llama_kv_cache_seq_add(ctx->kv_self, seq_id, p0, p1, delta);
-}
+    // number of cells moved
+    uint32_t n_moves = 0;
 
-void llama_kv_cache_seq_div(struct llama_context * ctx, llama_seq_id seq_id, llama_pos p0, llama_pos p1, int d) {
-    if (d == 1) {
-        return;
-    }
+    // each move requires 6*n_layer tensors (see build_defrag)
+    //   - source view, destination view, copy operation
+    //   - x2 for keys and values
+    //const uint32_t max_moves = llama_model_max_nodes(model)/(6*n_layer);
+    // TODO: tmp fix https://github.com/ggerganov/llama.cpp/issues/6685#issuecomment-2057579516
+    const uint32_t max_moves = (llama_model_max_nodes(lctx.model) - 2*n_layer)/(6*n_layer);
 
-    llama_kv_cache_seq_div(ctx->kv_self, seq_id, p0, p1, d);
-}
+    // determine which KV cells to move where
+    //
+    //  cell i moves to ids[i]
+    //
+    //  if ids[i] == i || ids[i] == n_kv, then cell i is not moved
+    //
+    std::vector ids(n_kv, n_kv);
 
-llama_pos llama_kv_cache_seq_pos_max(struct llama_context * ctx, llama_seq_id seq_id) {
-    return llama_kv_cache_seq_pos_max(ctx->kv_self, seq_id);
-}
+    for (uint32_t i0 = 0; i0 < n_used; ++i0) {
+        const auto & cell0 = kv_self.cells[i0];
 
-void llama_kv_cache_defrag(struct llama_context * ctx) {
-    llama_kv_cache_defrag(ctx->kv_self);
-}
+        if (!cell0.is_empty()) {
+            ids[i0] = i0;
 
-void llama_kv_cache_update(struct llama_context * ctx) {
-    llama_kv_cache_update_internal(*ctx);
-}
+            continue;
+        }
 
-bool llama_kv_cache_can_shift(struct llama_context * ctx) {
-    return !ctx->kv_self.recurrent && ctx->model.arch != LLM_ARCH_DEEPSEEK2; // not supported due to MLA
-}
+        // found a hole - fill it with data from the end of the cache
 
-// deprecated
-size_t llama_get_state_size(struct llama_context * ctx) {
-    return llama_state_get_size(ctx);
-}
+        uint32_t nh = 1;
 
-// deprecated
-size_t llama_copy_state_data(struct llama_context * ctx, uint8_t * dst) {
-    return llama_state_get_data(ctx, dst, -1);
-}
+        // determine the size of the hole
+        while (i0 + nh < n_used && kv_self.cells[i0 + nh].is_empty()) {
+            nh++;
+        }
 
-// deprecated
-size_t llama_set_state_data(struct llama_context * ctx, const uint8_t * src) {
-    return llama_state_set_data(ctx, src, -1);
-}
+        uint32_t nf = 0;
+        uint32_t is = n_kv - 1;
 
-// deprecated
-bool llama_load_session_file(struct llama_context * ctx, const char * path_session, llama_token * tokens_out, size_t n_token_capacity, size_t * n_token_count_out) {
-    return llama_state_load_file(ctx, path_session, tokens_out, n_token_capacity, n_token_count_out);
-}
+        // starting from the end, find nh non-empty cells
+        for (; is > i0; --is) {
+            const auto & cell1 = kv_self.cells[is];
 
-// deprecated
-bool llama_save_session_file(struct llama_context * ctx, const char * path_session, const llama_token * tokens, size_t n_token_count) {
-    return llama_state_save_file(ctx, path_session, tokens, n_token_count);
-}
+            if (cell1.is_empty() || ids[is] != n_kv) {
+                continue;
+            }
 
-// TODO: replace all non-fatal assertions with returned errors or exceptions
-struct llama_data_write {
-    virtual void write(const void * src, size_t size) = 0;
-    virtual void write_tensor_data(const struct ggml_tensor * tensor, size_t offset, size_t size) = 0;
-    virtual size_t get_size_written() = 0;
-    virtual ~llama_data_write() = default;
+            // non-empty cell which is not yet moved
+            nf++;
 
-    void write_string(const std::string & str) {
-        uint32_t str_size = str.size();
+            if (nf == nh) {
+                break;
+            }
+        }
 
-        write(&str_size,  sizeof(str_size));
-        write(str.data(), str_size);
-    }
+        // this can only happen if `n_used` is not accurate, which would be a bug
+        GGML_ASSERT(nf == nh && "KV defrag bug: nf != nh");
 
-    void write_model_info(const struct llama_context * ctx) {
-        std::string arch_str = LLM_ARCH_NAMES.at(ctx->model.arch);
-        write_string(arch_str);
-        // TODO: add more model-specific info which should prevent loading the session file if not identical
-    }
+        nf = 0;
 
-    //void write_rng(const std::mt19937 & rng) {
-    //    std::ostringstream rng_ss;
-    //    rng_ss << rng;
+        uint32_t i1 = is;
 
-    //    const std::string & rng_str = rng_ss.str();
+        // are we moving a continuous block of memory?
+        bool cont = false;
 
-    //    write_string(rng_str);
-    //}
+        // should we stop searching for the next move?
+        bool stop = false;
 
-    void write_output_ids(struct llama_context * ctx) {
-        llama_output_reorder(ctx);
+        // go back and move the nf cells to the hole
+        for (; i1 < n_kv; ++i1) {
+            auto & cell1 = kv_self.cells[i1];
 
-        const uint32_t n_outputs = ctx->n_outputs;
+            if (cell1.is_empty() || ids[i1] != n_kv) {
+                if (n_moves == max_moves) {
+                    stop = true;
+                    break;
+                }
 
-        std::vector output_pos;
+                cont = false;
+                continue;
+            }
 
-        const size_t    n_batch = ctx->cparams.n_batch;
-        const auto & output_ids = ctx->output_ids;
+            // this cell goes to (i0 + nf)
+            ids[i1] = i0 + nf;
 
-        GGML_ASSERT(n_outputs <= ctx->output_size);
+            // move the cell meta data
+            kv_self.cells[i0 + nf] = cell1;
 
-        output_pos.resize(n_outputs);
+            // clear the old cell and move the head there
+            cell1 = llama_kv_cell();
+            kv_self.head = n_used;
 
-        // build a more compact representation of the output ids
-        for (size_t i = 0; i < n_batch; ++i) {
-            // map an output id to a position in the batch
-            int32_t pos = output_ids[i];
-            if (pos >= 0) {
-                GGML_ASSERT((uint32_t) pos < n_outputs);
-                output_pos[pos] = i;
+            if (!cont) {
+                n_moves++;
+                cont = true;
             }
-        }
 
-        write(&n_outputs, sizeof(n_outputs));
+            nf++;
 
-        if (n_outputs) {
-            write(output_pos.data(), n_outputs * sizeof(int32_t));
+            if (nf == nh) {
+                break;
+            }
         }
-    }
 
-    void write_logits(const struct llama_context * ctx) {
-        const uint64_t logits_size = std::min((uint64_t) ctx->logits_size, (uint64_t) ctx->n_outputs * ctx->model.hparams.n_vocab);
-
-        write(&logits_size, sizeof(logits_size));
-
-        if (logits_size) {
-            write(ctx->logits, logits_size * sizeof(float));
+        if (stop || n_moves == max_moves) {
+            break;
         }
-    }
 
-    void write_embeddings(const struct llama_context * ctx) {
-        const uint64_t embeddings_size = std::min((uint64_t) ctx->embd_size, (uint64_t) ctx->n_outputs * ctx->model.hparams.n_embd);
+        //LLAMA_LOG_INFO("(tmp log) KV defrag: move [%u, %u) to [%u, %u)\n", is, i1 + 1, i0, i0 + nh);
 
-        write(&embeddings_size, sizeof(embeddings_size));
+        i0 += nh - 1;
+    }
 
-        if (embeddings_size) {
-            write(ctx->embd, embeddings_size * sizeof(float));
-        }
+    if (n_moves == 0) {
+        return;
     }
 
-    void write_kv_cache_meta(const llama_kv_cache & kv_self, const std::vector> & cell_ranges, llama_seq_id seq_id = -1) {
+    //LLAMA_LOG_INFO("(tmp log) KV defrag cell moves: %u\n", n_moves);
 
-        for (const auto & range : cell_ranges) {
-            for (uint32_t i = range.first; i < range.second; ++i) {
-                const auto & cell = kv_self.cells[i];
-                const llama_pos pos      = cell.pos;
-                const uint32_t  n_seq_id = seq_id == -1 ? cell.seq_id.size() : 0;
+    //LLAMA_LOG_INFO("expected gf nodes: %u\n", 6*n_moves*n_layer);
 
-                write(&pos,      sizeof(pos));
-                write(&n_seq_id, sizeof(n_seq_id));
+#if 0
+    // CPU defrag
+    //
+    // TODO: optimizations are possible:
+    //       - multiple threads
+    //       - avoid copying to the host memory when already there
+    //
+    // likely not worth the effort, as we have ggml_graph based defrag
+    //
 
-                if (n_seq_id) {
-                    for (auto seq_id : cell.seq_id) {
-                        write(&seq_id, sizeof(seq_id));
-                    }
-                }
-            }
-        }
-    }
+    const uint32_t n_embd_k_gqa = hparams.n_embd_k_gqa();
+    const uint32_t n_embd_v_gqa = hparams.n_embd_v_gqa();
 
-    void write_kv_cache_data(const struct llama_context * ctx, const std::vector> & cell_ranges) {
-        const struct llama_kv_cache & kv_self = ctx->kv_self;
-        const struct llama_hparams & hparams = ctx->model.hparams;
+    const uint32_t kv_size = kv_self.size;
 
-        const uint32_t v_trans = kv_self.v_trans ? 1 : 0;
-        const uint32_t n_layer = hparams.n_layer;
+    std::vector buf_k;
+    std::vector buf_v;
 
-        write(&v_trans, sizeof(v_trans));
-        write(&n_layer, sizeof(n_layer));
+    for (uint32_t il = 0; il < n_layer; ++il) {
+        const size_t k_size_row = ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa);
+        const size_t k_size     = ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa*kv_size);
 
-        std::vector tmp_buf;
+        const size_t v_size_el = ggml_type_size(kv_self.v_l[il]->type);
+        const size_t v_size    = ggml_row_size (kv_self.v_l[il]->type, n_embd_v_gqa*kv_size);
 
-        // Iterate and write all the keys first, each row is a cell
-        // Get whole range at a time
-        for (uint32_t il = 0; il < n_layer; ++il) {
-            const uint32_t n_embd_k_gqa = hparams.n_embd_k_gqa(il) + hparams.n_embd_k_s();
+        buf_k.resize(k_size);
+        buf_v.resize(v_size);
 
-            // Write key type
-            const int32_t k_type_i = (int32_t)kv_self.k_l[il]->type;
-            write(&k_type_i, sizeof(k_type_i));
+        ggml_backend_tensor_get(kv_self.k_l[il], buf_k.data(), 0, buf_k.size());
+        ggml_backend_tensor_get(kv_self.v_l[il], buf_v.data(), 0, buf_v.size());
 
-            // Write row size of key
-            const uint64_t k_size_row = ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa);
-            write(&k_size_row, sizeof(k_size_row));
+        // batch move [i, i+nm) to [id, id+nm)
+        // note: cells can move only to a lower index
+        for (uint32_t i = 0; i < n_kv; ++i) {
+            const uint32_t id = ids[i];
 
-            // Read each range of cells of k_size length each into tmp_buf and write out
-            for (const auto & range : cell_ranges) {
-                const size_t range_size = range.second - range.first;
-                const size_t buf_size = range_size * k_size_row;
-                write_tensor_data(kv_self.k_l[il], range.first * k_size_row, buf_size);
+            if (i == id || id == n_kv) {
+                continue;
             }
-        }
 
-        if (!kv_self.v_trans) {
-            for (uint32_t il = 0; il < n_layer; ++il) {
-                const uint32_t n_embd_v_gqa = hparams.n_embd_v_gqa(il) + hparams.n_embd_v_s();
-
-                // Write value type
-                const int32_t v_type_i = (int32_t)kv_self.v_l[il]->type;
-                write(&v_type_i, sizeof(v_type_i));
-
-                // Write row size of value
-                const uint64_t v_size_row = ggml_row_size(kv_self.v_l[il]->type, n_embd_v_gqa);
-                write(&v_size_row, sizeof(v_size_row));
+            uint32_t nm = 1;
 
-                // Read each range of cells of v_size length each into tmp_buf and write out
-                for (const auto & range : cell_ranges) {
-                    const size_t range_size = range.second - range.first;
-                    const size_t buf_size = range_size * v_size_row;
-                    write_tensor_data(kv_self.v_l[il], range.first * v_size_row, buf_size);
-                }
+            while (i + nm < n_kv && ids[i + nm] == id + nm) {
+                nm++;
             }
-        } else {
-            // When v is transposed, we also need the element size and get the element ranges from each row
-            const uint32_t kv_size = kv_self.size;
-            for (uint32_t il = 0; il < n_layer; ++il) {
-                const uint32_t n_embd_v_gqa = hparams.n_embd_v_gqa(il) + hparams.n_embd_v_s();
 
-                // Write value type
-                const int32_t v_type_i = (int32_t)kv_self.v_l[il]->type;
-                write(&v_type_i, sizeof(v_type_i));
+            // move keys
+            {
+                const int64_t os =  i*k_size_row;
+                const int64_t od = id*k_size_row;
 
-                // Write element size
-                const uint32_t v_size_el = ggml_type_size(kv_self.v_l[il]->type);
-                write(&v_size_el, sizeof(v_size_el));
+                memcpy(buf_k.data() + od, buf_k.data() + os, nm*k_size_row);
+            }
 
-                // Write GQA embedding size
-                write(&n_embd_v_gqa, sizeof(n_embd_v_gqa));
+            // move values (note: they are transposed)
+            {
+                const int64_t os =  i;
+                const int64_t od = id;
 
-                // For each row, we get the element values of each cell
                 for (uint32_t j = 0; j < n_embd_v_gqa; ++j) {
-                    // Read each range of cells of v_size_el length each into tmp_buf and write out
-                    for (const auto & range : cell_ranges) {
-                        const size_t range_size = range.second - range.first;
-                        const size_t src_offset = (range.first + j * kv_size) * v_size_el;
-                        const size_t buf_size = range_size * v_size_el;
-                        write_tensor_data(kv_self.v_l[il], src_offset, buf_size);
-                    }
+                    memcpy(buf_v.data() + (od + j*kv_size)*v_size_el, buf_v.data() + (os + j*kv_size)*v_size_el, nm*v_size_el);
                 }
             }
-        }
-    }
 
-    void write_kv_cache(const struct llama_context * ctx, llama_seq_id seq_id = -1) {
-        const struct llama_kv_cache & kv_self = ctx->kv_self;
-        std::vector> cell_ranges; // ranges, from inclusive, to exclusive
-        uint32_t cell_count = 0;
-
-        // Count the number of cells with the specified seq_id
-        // Find all the ranges of cells with this seq id (or all, when -1)
-        uint32_t cell_range_begin = kv_self.size;
-        for (uint32_t i = 0; i < kv_self.size; ++i) {
-            const auto & cell = kv_self.cells[i];
-            if ((seq_id == -1 && !cell.is_empty()) || cell.has_seq_id(seq_id)) {
-                ++cell_count;
-                if (cell_range_begin == kv_self.size) {
-                    cell_range_begin = i;
-                }
-            } else {
-                if (cell_range_begin != kv_self.size) {
-                    cell_ranges.emplace_back(cell_range_begin, i);
-                    cell_range_begin = kv_self.size;
-                }
-            }
-        }
-        if (cell_range_begin != kv_self.size) {
-            cell_ranges.emplace_back(cell_range_begin, kv_self.size);
+            i += nm - 1;
         }
 
-        // DEBUG CHECK: Sum of cell counts in ranges should equal the total cell count
-        uint32_t cell_count_check = 0;
-        for (const auto & range : cell_ranges) {
-            cell_count_check += range.second - range.first;
-        }
-        GGML_ASSERT(cell_count == cell_count_check);
+        ggml_backend_tensor_set(kv_self.k_l[il], buf_k.data(), 0, buf_k.size());
+        ggml_backend_tensor_set(kv_self.v_l[il], buf_v.data(), 0, buf_v.size());
+    }
+#else
+    // ggml_graph defrag
 
-        write(&cell_count, sizeof(cell_count));
+    ggml_backend_sched_reset(lctx.sched.get());
 
-        write_kv_cache_meta(kv_self, cell_ranges, seq_id);
-        write_kv_cache_data(ctx, cell_ranges);
-    }
-};
+    ggml_cgraph * gf = llama_build_graph_defrag(lctx, ids);
+
+    llama_graph_compute(lctx, gf, lctx.cparams.n_threads, lctx.threadpool);
+#endif
 
-struct llama_data_read {
-    virtual const uint8_t * read(size_t size) = 0;
-    virtual void read_to(void * dst, size_t size) = 0;
-    virtual size_t get_size_read() = 0;
-    virtual ~llama_data_read() = default;
+    //const int64_t t_end = ggml_time_us();
 
-    void read_string(std::string & str) {
-        uint32_t str_size;
-        read_to(&str_size, sizeof(str_size));
+    //LLAMA_LOG_INFO("(tmp log) KV defrag time: %.3f ms\n", (t_end - t_start)/1000.0);
+}
 
-        str.assign((const char *) read(str_size), str_size);
-    }
+static void llama_kv_cache_update_impl(struct llama_context & lctx) {
+    bool need_reserve = false;
 
-    // validate model information
-    void read_model_info(const struct llama_context * ctx) {
-        std::string cur_arch_str = LLM_ARCH_NAMES.at(ctx->model.arch);
-        std::string arch_str;
-        read_string(arch_str);
-        if (cur_arch_str != arch_str) {
-            throw std::runtime_error(format("wrong model arch: '%s' instead of '%s'", arch_str.c_str(), cur_arch_str.c_str()));
+    if (lctx.kv_self.has_shift) {
+        if (!llama_kv_cache_can_shift(&lctx)) {
+            GGML_ABORT("The current context does not support K-shift");
         }
-        // TODO: add more info which needs to be identical but which is not verified otherwise
-    }
 
-    //void read_rng(std::mt19937 & rng) {
-    //    std::string rng_str;
-    //    read_string(rng_str);
+        // apply K-shift if needed
+        if (lctx.model.hparams.rope_type != LLAMA_ROPE_TYPE_NONE) {
+            ggml_backend_sched_reset(lctx.sched.get());
 
-    //    std::istringstream rng_ss(rng_str);
-    //    rng_ss >> rng;
+            ggml_cgraph * gf = llama_build_graph_k_shift(lctx);
 
-    //    if (rng_ss.fail()) {
-    //        throw std::runtime_error("failed to load RNG state");
-    //    }
-    //}
+            ggml_backend_sched_alloc_graph(lctx.sched.get(), gf);
 
-    void read_output_ids(struct llama_context * ctx) {
-        std::vector output_pos;
+            llama_set_k_shift(lctx);
 
-        uint32_t n_outputs;
-        read_to(&n_outputs, sizeof(n_outputs));
+            llama_graph_compute(lctx, gf, lctx.cparams.n_threads, lctx.threadpool);
 
-        if (n_outputs > llama_output_reserve(*ctx, n_outputs)) {
-            throw std::runtime_error("could not reserve outputs");
+            need_reserve = true;
         }
 
-        if (n_outputs) {
-            output_pos.resize(n_outputs);
-            read_to(output_pos.data(), n_outputs * sizeof(int32_t));
+        {
+            auto & kv_self = lctx.kv_self;
 
-            for (int32_t i = 0; i < (int32_t) output_pos.size(); ++i) {
-                int32_t id = output_pos[i];
-                if ((uint32_t) id >= ctx->cparams.n_batch) {
-                    throw std::runtime_error(format("invalid output id, %d does not fit in batch size of %u", id, ctx->cparams.n_batch));
-                }
-                ctx->output_ids[id] = i;
-            }
+            kv_self.has_shift = false;
 
-            ctx->n_outputs = n_outputs;
+            for (uint32_t i = 0; i < kv_self.size; ++i) {
+                kv_self.cells[i].delta = 0;
+            }
         }
     }
 
-    void read_logits(struct llama_context * ctx) {
-        uint64_t logits_size;
-        read_to(&logits_size, sizeof(logits_size));
+    // defragment the KV cache if needed
+    if (lctx.kv_self.do_defrag) {
+        llama_kv_cache_defrag_impl(lctx);
 
-        if (ctx->logits_size < logits_size) {
-            throw std::runtime_error("logits buffer too small");
-        }
+        need_reserve = true;
 
-        if (logits_size) {
-            read_to(ctx->logits, logits_size * sizeof(float));
-        }
+        lctx.kv_self.do_defrag = false;
     }
 
-    void read_embeddings(struct llama_context * ctx) {
-        uint64_t embeddings_size;
-        read_to(&embeddings_size, sizeof(embeddings_size));
+    // reserve a worst case graph again
+    if (need_reserve) {
+        // TODO: extract to a function
+        // build worst-case graph
+        uint32_t n_seqs = 1; // TODO: worst-case number of sequences
+        uint32_t n_tokens = std::min(lctx.cparams.n_ctx, lctx.cparams.n_ubatch);
+        llama_token token = llama_token_bos(&lctx.model); // not actually used by llama_build_graph, but required to choose between token and embedding inputs graph
+        llama_ubatch ubatch = { true, n_tokens, n_tokens / n_seqs, n_seqs, &token, nullptr, nullptr, nullptr, nullptr, nullptr};
+        ggml_cgraph * gf = llama_build_graph(lctx, ubatch, true);
 
-        if (ctx->embd_size < embeddings_size) {
-            throw std::runtime_error("embeddings buffer too small");
+        // initialize scheduler with the worst-case graph
+        ggml_backend_sched_reset(lctx.sched.get());
+        if (!ggml_backend_sched_reserve(lctx.sched.get(), gf)) {
+            LLAMA_LOG_ERROR("%s: failed to allocate compute buffers\n", __func__);
         }
+    }
+}
 
-        if (embeddings_size) {
-            read_to(ctx->embd, embeddings_size * sizeof(float));
-        }
+int32_t llama_lora_adapter_set(
+            struct llama_context * ctx,
+            struct llama_lora_adapter * adapter,
+            float scale) {
+    ctx->lora_adapters[adapter] = scale;
+    return 0;
+}
+
+int32_t llama_lora_adapter_remove(
+            struct llama_context * ctx,
+            struct llama_lora_adapter * adapter) {
+    auto pos = ctx->lora_adapters.find(adapter);
+    if (pos != ctx->lora_adapters.end()) {
+        ctx->lora_adapters.erase(pos);
+        return 0;
     }
 
-    bool read_kv_cache_meta(struct llama_context * ctx, uint32_t cell_count, llama_seq_id dest_seq_id = -1) {
-        struct llama_kv_cache & kv_self = ctx->kv_self;
+    return -1;
+}
+
+void llama_lora_adapter_clear(struct llama_context * ctx) {
+    ctx->lora_adapters.clear();
+}
+
+// TODO: tmp
+int32_t llama_control_vector_apply(
+        struct llama_context * lctx,
+                 const float * data,
+                      size_t   len,
+                     int32_t   n_embd,
+                     int32_t   il_start,
+                     int32_t   il_end) {
+    return llama_control_vector_apply(lctx->cvec, lctx->model, data, len, n_embd, il_start, il_end);
+}
+
+//
+// interface implementation
+//
 
-        if (dest_seq_id != -1) {
-            // single sequence
+struct llama_context_params llama_context_default_params() {
+    struct llama_context_params result = {
+        /*.n_ctx                       =*/ 512,
+        /*.n_batch                     =*/ 2048,
+        /*.n_ubatch                    =*/ 512,
+        /*.n_seq_max                   =*/ 1,
+        /*.n_threads                   =*/ GGML_DEFAULT_N_THREADS, // TODO: better default
+        /*.n_threads_batch             =*/ GGML_DEFAULT_N_THREADS,
+        /*.rope_scaling_type           =*/ LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED,
+        /*.pooling_type                =*/ LLAMA_POOLING_TYPE_UNSPECIFIED,
+        /*.attention_type              =*/ LLAMA_ATTENTION_TYPE_UNSPECIFIED,
+        /*.rope_freq_base              =*/ 0.0f,
+        /*.rope_freq_scale             =*/ 0.0f,
+        /*.yarn_ext_factor             =*/ -1.0f,
+        /*.yarn_attn_factor            =*/ 1.0f,
+        /*.yarn_beta_fast              =*/ 32.0f,
+        /*.yarn_beta_slow              =*/ 1.0f,
+        /*.yarn_orig_ctx               =*/ 0,
+        /*.defrag_thold                =*/ -1.0f,
+        /*.cb_eval                     =*/ nullptr,
+        /*.cb_eval_user_data           =*/ nullptr,
+        /*.type_k                      =*/ GGML_TYPE_F16,
+        /*.type_v                      =*/ GGML_TYPE_F16,
+        /*.logits_all                  =*/ false,
+        /*.embeddings                  =*/ false,
+        /*.offload_kqv                 =*/ true,
+        /*.flash_attn                  =*/ false,
+        /*.no_perf                     =*/ true,
+        /*.abort_callback              =*/ nullptr,
+        /*.abort_callback_data         =*/ nullptr,
+    };
 
-            llama_kv_cache_seq_rm(kv_self, dest_seq_id, -1, -1);
+    return result;
+}
 
-            llama_ubatch batch = ctx->sbatch.reserve_ubatch(cell_count, /* has_embd */ false);
-            batch.n_tokens = cell_count;
-            batch.n_seq_tokens = cell_count;
-            batch.n_seqs = 1;
+struct llama_sampler_chain_params llama_sampler_chain_default_params() {
+    struct llama_sampler_chain_params result = {
+        /*.no_perf                     =*/ true,
+    };
 
-            for (uint32_t i = 0; i < cell_count; ++i) {
-                llama_pos pos;
-                uint32_t n_seq_id;
+    return result;
+}
 
-                read_to(&pos, sizeof(pos));
-                read_to(&n_seq_id, sizeof(n_seq_id));
+size_t llama_max_devices(void) {
+    return 16;
+}
 
-                if (n_seq_id != 0) {
-                    LLAMA_LOG_ERROR("%s: invalid seq_id-agnostic kv cell\n", __func__);
-                    return false;
-                }
+bool llama_supports_mmap(void) {
+    return llama_mmap::SUPPORTED;
+}
 
-                batch.pos[i] = pos;
-            }
-            batch.n_seq_id[0] = 1;
-            batch.seq_id[0] = &dest_seq_id;
-            if (!llama_kv_cache_find_slot(kv_self, batch)) {
-                LLAMA_LOG_ERROR("%s: failed to find available cells in kv cache\n", __func__);
-                return false;
-            }
+bool llama_supports_mlock(void) {
+    return llama_mlock::SUPPORTED;
+}
 
-            // DEBUG CHECK: kv_self.head should be our first cell, kv_self.head + cell_count - 1 should be our last cell (verify seq_id and pos values)
-            // Assume that this is one contiguous block of cells
-            GGML_ASSERT(kv_self.head + cell_count <= kv_self.size);
-            GGML_ASSERT(kv_self.cells[kv_self.head].pos == batch.pos[0]);
-            GGML_ASSERT(kv_self.cells[kv_self.head + cell_count - 1].pos == batch.pos[cell_count - 1]);
-            GGML_ASSERT(kv_self.cells[kv_self.head].has_seq_id(dest_seq_id));
-            GGML_ASSERT(kv_self.cells[kv_self.head + cell_count - 1].has_seq_id(dest_seq_id));
-        } else {
-            // whole KV cache restore
+bool llama_supports_gpu_offload(void) {
+    return ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_GPU) != nullptr ||
+           llama_supports_rpc();
+}
 
-            if (cell_count > kv_self.size) {
-                LLAMA_LOG_ERROR("%s: not enough cells in kv cache\n", __func__);
-                return false;
-            }
+bool llama_supports_rpc(void) {
+    return ggml_backend_reg_by_name("RPC") != nullptr;
+}
 
-            llama_kv_cache_clear(kv_self);
+void llama_backend_init(void) {
+    ggml_time_init();
 
-            for (uint32_t i = 0; i < cell_count; ++i) {
-                llama_kv_cell & cell = kv_self.cells[i];
+    // needed to initialize f16 tables
+    {
+        struct ggml_init_params params = { 0, NULL, false };
+        struct ggml_context * ctx = ggml_init(params);
+        ggml_free(ctx);
+    }
+}
 
-                llama_pos pos;
-                uint32_t  n_seq_id;
+void llama_numa_init(enum ggml_numa_strategy numa) {
+    if (numa != GGML_NUMA_STRATEGY_DISABLED) {
+        auto * dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
+        GGML_ASSERT(dev && "CPU backend is not loaded");
+        auto * reg = ggml_backend_dev_backend_reg(dev);
+        auto * numa_init_fn = (decltype(ggml_numa_init) *) ggml_backend_reg_get_proc_address(reg, "ggml_backend_cpu_numa_init");
+        numa_init_fn(numa);
+    }
+}
 
-                read_to(&pos,      sizeof(pos));
-                read_to(&n_seq_id, sizeof(n_seq_id));
+void llama_backend_free(void) {
+    ggml_quantize_free();
+}
 
-                cell.pos = pos;
+int64_t llama_time_us(void) {
+    return ggml_time_us();
+}
 
-                for (uint32_t j = 0; j < n_seq_id; ++j) {
-                    llama_seq_id seq_id;
-                    read_to(&seq_id, sizeof(seq_id));
+struct llama_model * llama_load_model_from_file(
+        const char * path_model,
+        struct llama_model_params params) {
+    return llama_model_load_from_file(path_model, params);
+}
 
-                    if (seq_id < 0 || (uint32_t) seq_id >= llama_n_seq_max(ctx)) {
-                        LLAMA_LOG_ERROR("%s: invalid seq_id, %d is out of range [0, %u)\n", __func__, seq_id, llama_n_seq_max(ctx));
-                        return false;
-                    }
+struct llama_model * llama_model_load_from_file(
+        const char * path_model,
+        struct llama_model_params params) {
+    ggml_time_init();
 
-                    cell.seq_id.insert(seq_id);
+    llama_model * model = new llama_model;
 
-                    if (kv_self.recurrent) {
-                        int32_t & tail = kv_self.cells[seq_id].tail;
-                        if (tail != -1) {
-                            LLAMA_LOG_ERROR("%s: duplicate tail for seq_id %d in cell %d and %d\n", __func__, seq_id, i, tail);
-                            return false;
-                        }
-                        tail = i;
-                    }
+    unsigned cur_percentage = 0;
+    if (params.progress_callback == NULL) {
+        params.progress_callback_user_data = &cur_percentage;
+        params.progress_callback = [](float progress, void * ctx) {
+            unsigned * cur_percentage_p = (unsigned *) ctx;
+            unsigned percentage = (unsigned) (100 * progress);
+            while (percentage > *cur_percentage_p) {
+                *cur_percentage_p = percentage;
+                LLAMA_LOG_CONT(".");
+                if (percentage >= 100) {
+                    LLAMA_LOG_CONT("\n");
                 }
             }
+            return true;
+        };
+    }
 
-            kv_self.head = 0;
-            kv_self.used = cell_count;
-        }
-
-        if (kv_self.recurrent) {
-            for (uint32_t i = 0; i < cell_count; ++i) {
-                uint32_t cell_id = kv_self.head + i;
-                // make sure the recurrent states will keep their restored state
-                kv_self.cells[cell_id].src = cell_id;
-            }
+    if (params.rpc_servers != nullptr && params.rpc_servers[0] != '\0') {
+        // split the servers set them into model->rpc_servers
+        std::string servers(params.rpc_servers);
+        size_t pos = 0;
+        while ((pos = servers.find(',')) != std::string::npos) {
+            std::string server = servers.substr(0, pos);
+            model->rpc_servers.push_back(server);
+            servers.erase(0, pos + 1);
         }
-
-        return true;
+        model->rpc_servers.push_back(servers);
     }
 
-    bool read_kv_cache_data(struct llama_context * ctx, uint32_t cell_count) {
-        const struct llama_hparams & hparams = ctx->model.hparams;
-        struct llama_kv_cache & kv_self = ctx->kv_self;
-        uint32_t v_trans;
-        uint32_t n_layer;
-        read_to(&v_trans, sizeof(v_trans));
-        read_to(&n_layer, sizeof(n_layer));
-
-        if (n_layer != hparams.n_layer) {
-            LLAMA_LOG_ERROR("%s: mismatched layer count (%u instead of %u)\n", __func__, n_layer, hparams.n_layer);
-            return false;
-        }
-        if (cell_count > kv_self.size) {
-            LLAMA_LOG_ERROR("%s: not enough cells in kv cache to restore state (%u > %u)\n", __func__, cell_count, kv_self.size);
-            return false;
-        }
-        if (kv_self.v_trans != (bool) v_trans) {
-            LLAMA_LOG_ERROR("%s: incompatible V transposition\n", __func__);
-            return false;
+    // add RPC devices
+    if (!model->rpc_servers.empty()) {
+        ggml_backend_reg_t rpc_reg = ggml_backend_reg_by_name("RPC");
+        if (!rpc_reg) {
+            LLAMA_LOG_ERROR("%s: failed to find RPC backend\n", __func__);
+            llama_model_free(model);
+            return nullptr;
         }
 
-        // For each layer, read the keys for each cell, one row is one cell, read as one contiguous block
-        for (uint32_t il = 0; il < n_layer; ++il) {
-            const uint32_t n_embd_k_gqa = hparams.n_embd_k_gqa(il) + hparams.n_embd_k_s();
-
-            // Read type of key
-            int32_t k_type_i_ref;
-            read_to(&k_type_i_ref, sizeof(k_type_i_ref));
-            const int32_t k_type_i = (int32_t)kv_self.k_l[il]->type;
-            if (k_type_i != k_type_i_ref) {
-                LLAMA_LOG_ERROR("%s: mismatched key type (%d != %d, layer %d)\n", __func__, k_type_i, k_type_i_ref, il);
-                return false;
-            }
-
-            // Read row size of key
-            uint64_t k_size_row_ref;
-            read_to(&k_size_row_ref, sizeof(k_size_row_ref));
-            const size_t k_size_row = ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa);
-            if (k_size_row != k_size_row_ref) {
-                LLAMA_LOG_ERROR("%s: mismatched key row size (%zu != %zu, layer %d)\n", __func__, k_size_row, (size_t) k_size_row_ref, il);
-                return false;
-            }
-
-            if (cell_count) {
-                // Read and set the keys for the whole cell range
-                ggml_backend_tensor_set(kv_self.k_l[il], read(cell_count * k_size_row), kv_self.head * k_size_row, cell_count * k_size_row);
-            }
+        typedef ggml_backend_dev_t (*ggml_backend_rpc_add_device_t)(const char * endpoint);
+        ggml_backend_rpc_add_device_t ggml_backend_rpc_add_device_fn = (ggml_backend_rpc_add_device_t) ggml_backend_reg_get_proc_address(rpc_reg, "ggml_backend_rpc_add_device");
+        if (!ggml_backend_rpc_add_device_fn) {
+            LLAMA_LOG_ERROR("%s: failed to find RPC device add function\n", __func__);
+            llama_model_free(model);
+            return nullptr;
         }
 
-        if (!kv_self.v_trans) {
-            for (uint32_t il = 0; il < n_layer; ++il) {
-                const uint32_t n_embd_v_gqa = hparams.n_embd_v_gqa(il) + hparams.n_embd_v_s();
-
-                // Read type of value
-                int32_t v_type_i_ref;
-                read_to(&v_type_i_ref, sizeof(v_type_i_ref));
-                const int32_t v_type_i = (int32_t)kv_self.v_l[il]->type;
-                if (v_type_i != v_type_i_ref) {
-                    LLAMA_LOG_ERROR("%s: mismatched value type (%d != %d, layer %d)\n", __func__, v_type_i, v_type_i_ref, il);
-                    return false;
-                }
-
-                // Read row size of value
-                uint64_t v_size_row_ref;
-                read_to(&v_size_row_ref, sizeof(v_size_row_ref));
-                const size_t v_size_row = ggml_row_size(kv_self.v_l[il]->type, n_embd_v_gqa);
-                if (v_size_row != v_size_row_ref) {
-                    LLAMA_LOG_ERROR("%s: mismatched value row size (%zu != %zu, layer %d)\n", __func__, v_size_row, (size_t) v_size_row_ref, il);
-                    return false;
-                }
-
-                if (cell_count) {
-                    // Read and set the values for the whole cell range
-                    ggml_backend_tensor_set(kv_self.v_l[il], read(cell_count * v_size_row), kv_self.head * v_size_row, cell_count * v_size_row);
-                }
+        for (const std::string & server : model->rpc_servers) {
+            ggml_backend_dev_t dev = ggml_backend_rpc_add_device_fn(server.c_str());
+            if (dev) {
+                model->devices.push_back(dev);
+            } else {
+                LLAMA_LOG_ERROR("%s: failed to add RPC device for server '%s'\n", __func__, server.c_str());
+                llama_model_free(model);
+                return nullptr;
             }
-        } else {
-            // For each layer, read the values for each cell (transposed)
-            for (uint32_t il = 0; il < n_layer; ++il) {
-                const uint32_t n_embd_v_gqa = hparams.n_embd_v_gqa(il) + hparams.n_embd_v_s();
-
-                // Read type of value
-                int32_t v_type_i_ref;
-                read_to(&v_type_i_ref, sizeof(v_type_i_ref));
-                const int32_t v_type_i = (int32_t)kv_self.v_l[il]->type;
-                if (v_type_i != v_type_i_ref) {
-                    LLAMA_LOG_ERROR("%s: mismatched value type (%d != %d, layer %d)\n", __func__, v_type_i, v_type_i_ref, il);
-                    return false;
-                }
-
-                // Read element size of value
-                uint32_t v_size_el_ref;
-                read_to(&v_size_el_ref, sizeof(v_size_el_ref));
-                const size_t v_size_el = ggml_type_size(kv_self.v_l[il]->type);
-                if (v_size_el != v_size_el_ref) {
-                    LLAMA_LOG_ERROR("%s: mismatched value element size (%zu != %zu, layer %d)\n", __func__, v_size_el, (size_t) v_size_el_ref, il);
-                    return false;
-                }
+        }
+    }
 
-                // Read GQA embedding size
-                uint32_t n_embd_v_gqa_ref;
-                read_to(&n_embd_v_gqa_ref, sizeof(n_embd_v_gqa_ref));
-                if (n_embd_v_gqa != n_embd_v_gqa_ref) {
-                    LLAMA_LOG_ERROR("%s: mismatched GQA embedding size (%u != %u, layer %d)\n", __func__, n_embd_v_gqa, n_embd_v_gqa_ref, il);
-                    return false;
-                }
+    // create list of devices to use with this model
+    if (params.devices) {
+        for (ggml_backend_dev_t * dev = params.devices; *dev; ++dev) {
+            model->devices.push_back(*dev);
+        }
+    } else {
+        // use all available devices
+        for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
+            ggml_backend_dev_t dev = ggml_backend_dev_get(i);
+            switch (ggml_backend_dev_type(dev)) {
+                case GGML_BACKEND_DEVICE_TYPE_CPU:
+                case GGML_BACKEND_DEVICE_TYPE_ACCEL:
+                    // skip CPU backends since they are handled separately
+                    break;
 
-                if (cell_count) {
-                    // For each row in the transposed matrix, read the values for the whole cell range
-                    for (uint32_t j = 0; j < n_embd_v_gqa; ++j) {
-                        const size_t dst_offset = (kv_self.head + j * kv_self.size) * v_size_el;
-                        ggml_backend_tensor_set(kv_self.v_l[il], read(cell_count * v_size_el), dst_offset, cell_count * v_size_el);
-                    }
-                }
+                case GGML_BACKEND_DEVICE_TYPE_GPU:
+                    model->devices.push_back(dev);
+                    break;
             }
         }
-        return true;
     }
 
-    void read_kv_cache(struct llama_context * ctx, llama_seq_id seq_id = -1) {
-        uint32_t cell_count;
-        read_to(&cell_count, sizeof(cell_count));
+    // if using single GPU mode, remove all except the main GPU
+    if (params.split_mode == LLAMA_SPLIT_MODE_NONE) {
+        if (params.main_gpu < 0 || params.main_gpu >= (int)model->devices.size()) {
+            LLAMA_LOG_ERROR("%s: invalid value for main_gpu: %d (available devices: %d)\n", __func__, params.main_gpu, (int)model->devices.size());
+            llama_model_free(model);
+            return nullptr;
+        }
+        ggml_backend_dev_t main_gpu = model->devices[params.main_gpu];
+        model->devices.clear();
+        model->devices.push_back(main_gpu);
+    }
 
-        bool res = read_kv_cache_meta(ctx, cell_count, seq_id) && read_kv_cache_data(ctx, cell_count);
+    for (auto * dev : model->devices) {
+        size_t free, total; // NOLINT
+        ggml_backend_dev_memory(dev, &free, &total);
+        LLAMA_LOG_INFO("%s: using device %s (%s) - %zu MiB free\n", __func__, ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), free/1024/1024);
+    }
 
-        if (!res) {
-            if (seq_id == -1) {
-                llama_kv_cache_clear(ctx);
-            } else {
-                llama_kv_cache_seq_rm(ctx, seq_id, -1, -1);
-            }
-            throw std::runtime_error("failed to restore kv cache");
+    int status = llama_model_load(path_model, *model, params);
+    GGML_ASSERT(status <= 0);
+    if (status < 0) {
+        if (status == -1) {
+            LLAMA_LOG_ERROR("%s: failed to load model\n", __func__);
+        } else if (status == -2) {
+            LLAMA_LOG_INFO("%s: cancelled model load\n", __func__);
         }
+
+        llama_model_free(model);
+        return nullptr;
     }
-};
 
-struct llama_data_write_dummy : llama_data_write {
-    size_t size_written = 0;
+    return model;
+}
 
-    llama_data_write_dummy() {}
+struct llama_context * llama_new_context_with_model(
+                 struct llama_model * model,
+        struct llama_context_params   params) {
 
-    void write(const void * /* src */, size_t size) override {
-        size_written += size;
+    if (!model) {
+        LLAMA_LOG_ERROR("%s: model cannot be NULL\n", __func__);
+        return nullptr;
     }
 
-    void write_tensor_data(const struct ggml_tensor * /* tensor */, size_t /* offset */, size_t size) override {
-        size_written += size;
+    if (params.n_batch == 0 && params.n_ubatch == 0) {
+        LLAMA_LOG_ERROR("%s: n_batch and n_ubatch cannot both be zero\n", __func__);
+        return nullptr;
     }
 
-    size_t get_size_written() override {
-        return size_written;
+    if (params.n_ctx == 0 && model->hparams.n_ctx_train == 0) {
+        LLAMA_LOG_ERROR("%s: n_ctx and model->hparams.n_ctx_train cannot both be zero\n", __func__);
+        return nullptr;
     }
-};
-
-struct llama_data_write_buffer : llama_data_write {
-    uint8_t * ptr;
-    size_t buf_size = 0;
-    size_t size_written = 0;
 
-    llama_data_write_buffer(uint8_t * p, size_t len) : ptr(p), buf_size(len) {}
-
-    void write(const void * src, size_t size) override {
-        if (size > buf_size) {
-            throw std::runtime_error("unexpectedly reached end of buffer");
-        }
-        memcpy(ptr, src, size);
-        ptr += size;
-        size_written += size;
-        buf_size -= size;
+    if (params.flash_attn && model->arch == LLM_ARCH_GROK) {
+        LLAMA_LOG_WARN("%s: flash_attn is not compatible with Grok - forcing off\n", __func__);
+        params.flash_attn = false;
     }
 
-    void write_tensor_data(const struct ggml_tensor * tensor, size_t offset, size_t size) override {
-        if (size > buf_size) {
-            throw std::runtime_error("unexpectedly reached end of buffer");
-        }
-        ggml_backend_tensor_get(tensor, ptr, offset, size);
-        ptr += size;
-        size_written += size;
-        buf_size -= size;
+    if (params.flash_attn && model->hparams.n_embd_head_k != model->hparams.n_embd_head_v) {
+        LLAMA_LOG_WARN("%s: flash_attn requires n_embd_head_k == n_embd_head_v - forcing off\n", __func__);
+        params.flash_attn = false;
     }
 
-    size_t get_size_written() override {
-        return size_written;
+    if (ggml_is_quantized(params.type_v) && !params.flash_attn) {
+        LLAMA_LOG_ERROR("%s: V cache quantization requires flash_attn\n", __func__);
+        return nullptr;
     }
-};
 
-struct llama_data_read_buffer : llama_data_read {
-    const uint8_t * ptr;
-    size_t buf_size = 0;
-    size_t size_read = 0;
+    llama_context * ctx = new llama_context(*model);
 
-    llama_data_read_buffer(const uint8_t * p, size_t len) : ptr(p), buf_size(len) {}
+    const auto & hparams = model->hparams;
+    auto       & cparams = ctx->cparams;
 
-    const uint8_t * read(size_t size) override {
-        const uint8_t * base_ptr = ptr;
-        if (size > buf_size) {
-            throw std::runtime_error("unexpectedly reached end of buffer");
-        }
-        ptr += size;
-        size_read += size;
-        buf_size -= size;
-        return base_ptr;
-    }
+    cparams.n_seq_max        = std::max(1u, params.n_seq_max);
+    cparams.n_threads        = params.n_threads;
+    cparams.n_threads_batch  = params.n_threads_batch;
+    cparams.yarn_ext_factor  = params.yarn_ext_factor;
+    cparams.yarn_attn_factor = params.yarn_attn_factor;
+    cparams.yarn_beta_fast   = params.yarn_beta_fast;
+    cparams.yarn_beta_slow   = params.yarn_beta_slow;
+    cparams.defrag_thold     = params.defrag_thold;
+    cparams.embeddings       = params.embeddings;
+    cparams.offload_kqv      = params.offload_kqv;
+    cparams.flash_attn       = params.flash_attn;
+    cparams.no_perf          = params.no_perf;
+    cparams.pooling_type     = params.pooling_type;
 
-    void read_to(void * dst, size_t size) override {
-        memcpy(dst, read(size), size);
-    }
+    cparams.n_ctx            = params.n_ctx           == 0    ? hparams.n_ctx_train           : params.n_ctx;
+    cparams.rope_freq_base   = params.rope_freq_base  == 0.0f ? hparams.rope_freq_base_train  : params.rope_freq_base;
+    cparams.rope_freq_scale  = params.rope_freq_scale == 0.0f ? hparams.rope_freq_scale_train : params.rope_freq_scale;
+
+    // this is necessary due to kv_self.n being padded later during inference
+    cparams.n_ctx            = GGML_PAD(cparams.n_ctx, llama_kv_cache_get_padding(cparams));
+
+    // with causal attention, the batch size is limited by the context size
+    cparams.n_batch          = hparams.causal_attn ? std::min(cparams.n_ctx, params.n_batch) : params.n_batch;
 
-    size_t get_size_read() override {
-        return size_read;
+    // the batch has to be at least GGML_KQ_MASK_PAD because we will be padding the KQ_mask
+    // this is required by GPU kernels in order to avoid out-of-bounds accesses (e.g. ggml_flash_attn_ext)
+    // ref: https://github.com/ggerganov/llama.cpp/pull/5021
+    if (cparams.n_batch < GGML_KQ_MASK_PAD) {
+        LLAMA_LOG_WARN("%s: n_batch is less than GGML_KQ_MASK_PAD - increasing to %d\n", __func__, GGML_KQ_MASK_PAD);
+        cparams.n_batch = GGML_KQ_MASK_PAD;
     }
-};
 
-struct llama_data_write_file : llama_data_write {
-    llama_file * file;
-    size_t size_written = 0;
-    std::vector temp_buffer;
+    cparams.n_ubatch         = std::min(cparams.n_batch, params.n_ubatch == 0 ? params.n_batch : params.n_ubatch);
 
-    llama_data_write_file(llama_file * f) : file(f) {}
+    cparams.n_ctx_orig_yarn  = params.yarn_orig_ctx    != 0 ? params.yarn_orig_ctx    :
+                               hparams.n_ctx_orig_yarn != 0 ? hparams.n_ctx_orig_yarn :
+                                                              hparams.n_ctx_train;
 
-    void write(const void * src, size_t size) override {
-        file->write_raw(src, size);
-        size_written += size;
-    }
+    cparams.cb_eval           = params.cb_eval;
+    cparams.cb_eval_user_data = params.cb_eval_user_data;
 
-    void write_tensor_data(const struct ggml_tensor * tensor, size_t offset, size_t size) override {
-        temp_buffer.resize(size);
-        ggml_backend_tensor_get(tensor, temp_buffer.data(), offset, size);
-        write(temp_buffer.data(), temp_buffer.size());
+    auto rope_scaling_type = params.rope_scaling_type;
+    if (rope_scaling_type == LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED) {
+        rope_scaling_type = hparams.rope_scaling_type_train;
     }
 
-    size_t get_size_written() override {
-        return size_written;
+    if (rope_scaling_type == LLAMA_ROPE_SCALING_TYPE_NONE) {
+        cparams.rope_freq_scale = 1.0f; // never scale if scaling type is none
     }
-};
 
-struct llama_data_read_file : llama_data_read {
-    llama_file * file;
-    size_t size_read = 0;
-    std::vector temp_buffer;
+    if (cparams.yarn_ext_factor < 0.0f) { // negative indicates 'not set'
+        cparams.yarn_ext_factor = rope_scaling_type == LLAMA_ROPE_SCALING_TYPE_YARN ? 1.0f : 0.0f;
+    }
 
-    llama_data_read_file(llama_file * f) : file(f) {}
+    cparams.yarn_attn_factor *= hparams.rope_attn_factor;
 
-    void read_to(void * dst, size_t size) override {
-        file->read_raw(dst, size);
-        size_read += size;
+    if (cparams.pooling_type == LLAMA_POOLING_TYPE_UNSPECIFIED) {
+        if (hparams.pooling_type == LLAMA_POOLING_TYPE_UNSPECIFIED) {
+            cparams.pooling_type = LLAMA_POOLING_TYPE_NONE;
+        } else {
+            cparams.pooling_type = hparams.pooling_type;
+        }
     }
 
-    const uint8_t * read(size_t size) override {
-        temp_buffer.resize(size);
-        read_to(temp_buffer.data(), size);
-        return temp_buffer.data();
+    if (params.attention_type == LLAMA_ATTENTION_TYPE_UNSPECIFIED) {
+        cparams.causal_attn = hparams.causal_attn;
+    } else {
+        cparams.causal_attn = params.attention_type == LLAMA_ATTENTION_TYPE_CAUSAL;
     }
 
-    size_t get_size_read() override {
-        return size_read;
-    }
-};
+    const uint32_t n_ctx_per_seq = cparams.n_ctx / cparams.n_seq_max;
 
-/** copy state data into either a buffer or file depending on the passed in context
- *
- * file context:
- * llama_file file("/path", "wb");
- * llama_data_write_file data_ctx(&file);
- * llama_state_get_data_internal(ctx, data_ctx);
- *
- * buffer context:
- * std::vector buf(max_size, 0);
- * llama_data_write_buffer data_ctx(buf.data(), max_size);
- * llama_state_get_data_internal(ctx, data_ctx);
- *
-*/
-static size_t llama_state_get_data_internal(struct llama_context * ctx, llama_data_write & data_ctx) {
-    llama_synchronize(ctx);
-
-    data_ctx.write_model_info(ctx);
-
-    // copy outputs
-    data_ctx.write_output_ids(ctx);
-    data_ctx.write_logits(ctx);
-    data_ctx.write_embeddings(ctx);
-
-    data_ctx.write_kv_cache(ctx);
-
-    return data_ctx.get_size_written();
-}
+    LLAMA_LOG_INFO("%s: n_seq_max     = %u\n",   __func__, cparams.n_seq_max);
+    LLAMA_LOG_INFO("%s: n_ctx         = %u\n",   __func__, cparams.n_ctx);
+    LLAMA_LOG_INFO("%s: n_ctx_per_seq = %u\n",   __func__, n_ctx_per_seq);
+    LLAMA_LOG_INFO("%s: n_batch       = %u\n",   __func__, cparams.n_batch);
+    LLAMA_LOG_INFO("%s: n_ubatch      = %u\n",   __func__, cparams.n_ubatch);
+    LLAMA_LOG_INFO("%s: flash_attn    = %d\n",   __func__, cparams.flash_attn);
+    LLAMA_LOG_INFO("%s: freq_base     = %.1f\n", __func__, cparams.rope_freq_base);
+    LLAMA_LOG_INFO("%s: freq_scale    = %g\n",   __func__, cparams.rope_freq_scale);
 
-size_t llama_state_get_data(struct llama_context * ctx, uint8_t * dst, size_t size) {
-    llama_data_write_buffer data_ctx(dst, size);
-    try {
-        return llama_state_get_data_internal(ctx, data_ctx);
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: error saving state: %s\n", __func__, err.what());
-        return 0;
+    if (n_ctx_per_seq < hparams.n_ctx_train) {
+        LLAMA_LOG_WARN("%s: n_ctx_per_seq (%u) < n_ctx_train (%u) -- the full capacity of the model will not be utilized\n",
+                __func__, n_ctx_per_seq, hparams.n_ctx_train);
     }
-}
 
-// Returns the *actual* size of the state.
-// Intended to be used when saving to state to a buffer.
-size_t llama_state_get_size(struct llama_context * ctx) {
-    llama_data_write_dummy data_ctx;
-    try {
-        return llama_state_get_data_internal(ctx, data_ctx);
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: error getting state size: %s\n", __func__, err.what());
-        return 0;
+    if (n_ctx_per_seq > hparams.n_ctx_train) {
+        LLAMA_LOG_WARN("%s: n_ctx_pre_seq (%u) > n_ctx_train (%u) -- possible training context overflow\n",
+                __func__, n_ctx_per_seq, hparams.n_ctx_train);
     }
-}
-
-static size_t llama_state_set_data_internal(struct llama_context * ctx, llama_data_read & data_ctx) {
-    llama_synchronize(ctx);
 
-    data_ctx.read_model_info(ctx);
-
-    // set outputs
-    data_ctx.read_output_ids(ctx);
-    data_ctx.read_logits(ctx);
-    data_ctx.read_embeddings(ctx);
+    ctx->logits_all = params.logits_all;
 
-    data_ctx.read_kv_cache(ctx);
+    // build worst-case graph for encoder if a model contains encoder
+    ctx->is_encoding = llama_model_has_encoder(model);
 
-    return data_ctx.get_size_read();
-}
+    uint32_t kv_size = cparams.n_ctx;
+    ggml_type type_k = params.type_k;
+    ggml_type type_v = params.type_v;
 
-// Sets the state reading from the specified source address
-size_t llama_state_set_data(struct llama_context * ctx, const uint8_t * src, size_t size) {
-    llama_data_read_buffer data_ctx(src, size);
-    try {
-        return llama_state_set_data_internal(ctx, data_ctx);
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: error loading state: %s\n", __func__, err.what());
-        return 0;
+    // Mamba only needs a constant number of KV cache cells per sequence
+    if (llama_model_is_recurrent(model)) {
+        // Mamba needs at least as many KV cells as there are sequences kept at any time
+        kv_size = std::max((uint32_t) 1, params.n_seq_max);
+        // it's probably best to keep as much precision as possible for the states
+        type_k = GGML_TYPE_F32; // required by ggml_ssm_conv for Mamba's conv_states
+        type_v = GGML_TYPE_F32; // required by ggml_ssm_scan for Mamba's ssm_states
     }
-}
-
-static bool llama_state_load_file_internal(struct llama_context * ctx, const char * path_session, llama_token * tokens_out, size_t n_token_capacity, size_t * n_token_count_out) {
-    llama_file file(path_session, "rb");
 
-    // sanity checks
-    {
-        const uint32_t magic   = file.read_u32();
-        const uint32_t version = file.read_u32();
+    GGML_ASSERT(hparams.n_embd_head_k % ggml_blck_size(type_k) == 0);
+    GGML_ASSERT(hparams.n_embd_head_v % ggml_blck_size(type_v) == 0);
 
-        if (magic != LLAMA_SESSION_MAGIC || version != LLAMA_SESSION_VERSION) {
-            LLAMA_LOG_ERROR("%s: unknown (magic, version) for session file: %08x, %08x\n", __func__, magic, version);
-            return false;
+    if (!hparams.vocab_only) {
+        // GPU backends
+        for (auto * dev : model->devices) {
+            ggml_backend_t backend = ggml_backend_dev_init(dev, nullptr);
+            if (backend == nullptr) {
+                LLAMA_LOG_ERROR("%s: failed to initialize %s backend\n", __func__, ggml_backend_dev_name(dev));
+                llama_free(ctx);
+                return nullptr;
+            }
+            ctx->backends.emplace_back(backend);
         }
-    }
-
-    // load the prompt
-    {
-        const uint32_t n_token_count = file.read_u32();
 
-        if (n_token_count > n_token_capacity) {
-            LLAMA_LOG_ERROR("%s: token count in session file exceeded capacity! %u > %zu\n", __func__, n_token_count, n_token_capacity);
-            return false;
+        // add ACCEL backends (such as BLAS)
+        for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
+            ggml_backend_dev_t dev = ggml_backend_dev_get(i);
+            if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_ACCEL) {
+                ggml_backend_t backend = ggml_backend_dev_init(dev, nullptr);
+                if (backend == nullptr) {
+                    LLAMA_LOG_ERROR("%s: failed to initialize %s backend\n", __func__, ggml_backend_dev_name(dev));
+                    llama_free(ctx);
+                    return nullptr;
+                }
+                ctx->backends.emplace_back(backend);
+            }
         }
 
-        file.read_raw(tokens_out, sizeof(llama_token) * n_token_count);
-        *n_token_count_out = n_token_count;
-    }
-
-    // restore the context state
-    {
-        const size_t n_state_size_cur = file.size - file.tell();
-
-        llama_data_read_file data_ctx(&file);
-        const size_t n_read = llama_state_set_data_internal(ctx, data_ctx);
-
-        if (n_read != n_state_size_cur) {
-            LLAMA_LOG_ERROR("%s: did not read all of the session file data! size %zu, got %zu\n", __func__, n_state_size_cur, n_read);
-            return false;
+        // add CPU backend
+        ctx->backend_cpu = ggml_backend_init_by_type(GGML_BACKEND_DEVICE_TYPE_CPU, nullptr);
+        if (ctx->backend_cpu == nullptr) {
+            LLAMA_LOG_ERROR("%s: failed to initialize CPU backend\n", __func__);
+            llama_free(ctx);
+            return nullptr;
         }
-    }
-    return true;
-}
-
-bool llama_state_load_file(struct llama_context * ctx, const char * path_session, llama_token * tokens_out, size_t n_token_capacity, size_t * n_token_count_out) {
-    try {
-        return llama_state_load_file_internal(ctx, path_session, tokens_out, n_token_capacity, n_token_count_out);
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: error loading session file: %s\n", __func__, err.what());
-        return false;
-    }
-}
+        ctx->backends.emplace_back(ctx->backend_cpu);
 
-static bool llama_state_save_file_internal(struct llama_context * ctx, const char * path_session, const llama_token * tokens, size_t n_token_count) {
-    llama_file file(path_session, "wb");
+        // create a list of the set_n_threads functions in the backends
+        for (auto & backend : ctx->backends) {
+            ggml_backend_dev_t dev = ggml_backend_get_device(backend.get());
+            ggml_backend_reg_t reg = dev ? ggml_backend_dev_backend_reg(dev) : nullptr;
+            if (reg) {
+                auto ggml_backend_set_n_threads_fn = (ggml_backend_set_n_threads_t) ggml_backend_reg_get_proc_address(reg, "ggml_backend_set_n_threads");
+                if (ggml_backend_set_n_threads_fn) {
+                    ctx->set_n_threads_fns.emplace_back(backend.get(), ggml_backend_set_n_threads_fn);
+                }
+            }
+        }
 
-    file.write_u32(LLAMA_SESSION_MAGIC);
-    file.write_u32(LLAMA_SESSION_VERSION);
+        llama_set_abort_callback(ctx, params.abort_callback, params.abort_callback_data);
 
-    // save the prompt
-    file.write_u32((uint32_t) n_token_count);
-    file.write_raw(tokens, sizeof(llama_token) * n_token_count);
+        if (!llama_kv_cache_init(ctx->kv_self, ctx->model, ctx->cparams, type_k, type_v, kv_size, cparams.offload_kqv)) {
+            LLAMA_LOG_ERROR("%s: llama_kv_cache_init() failed for self-attention cache\n", __func__);
+            llama_free(ctx);
+            return nullptr;
+        }
 
-    // save the context state using stream saving
-    llama_data_write_file data_ctx(&file);
-    llama_state_get_data_internal(ctx, data_ctx);
+        {
+            size_t memory_size_k = 0;
+            size_t memory_size_v = 0;
 
-    return true;
-}
+            for (auto & k : ctx->kv_self.k_l) {
+                memory_size_k += ggml_nbytes(k);
+            }
 
-bool llama_state_save_file(struct llama_context * ctx, const char * path_session, const llama_token * tokens, size_t n_token_count) {
-    try {
-        return llama_state_save_file_internal(ctx, path_session, tokens, n_token_count);
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: error saving session file: %s\n", __func__, err.what());
-        return false;
-    }
-}
+            for (auto & v : ctx->kv_self.v_l) {
+                memory_size_v += ggml_nbytes(v);
+            }
 
-static size_t llama_state_seq_get_data_internal(struct llama_context * ctx, llama_data_write & data_ctx, llama_seq_id seq_id) {
-    llama_synchronize(ctx);
+            LLAMA_LOG_INFO("%s: KV self size  = %7.2f MiB, K (%s): %7.2f MiB, V (%s): %7.2f MiB\n", __func__,
+                      (float)(memory_size_k + memory_size_v) / (1024.0f * 1024.0f),
+                ggml_type_name(type_k), (float)memory_size_k / (1024.0f * 1024.0f),
+                ggml_type_name(type_v), (float)memory_size_v / (1024.0f * 1024.0f));
+        }
 
-    data_ctx.write_kv_cache(ctx, seq_id);
+        // graph outputs buffer
+        {
+            // resized during inference when a batch uses more outputs
+            if (llama_output_reserve(*ctx, params.n_seq_max) < params.n_seq_max) {
+                LLAMA_LOG_ERROR("%s: failed to reserve initial output buffer\n", __func__);
+                llama_free(ctx);
+                return nullptr;
+            }
 
-    return data_ctx.get_size_written();
-}
+            LLAMA_LOG_INFO("%s: %10s  output buffer size = %8.2f MiB\n", __func__,
+                    ggml_backend_buffer_name(ctx->buf_output.get()),
+                    ggml_backend_buffer_get_size(ctx->buf_output.get()) / 1024.0 / 1024.0);
+        }
 
-size_t llama_state_seq_get_size(struct llama_context * ctx, llama_seq_id seq_id) {
-    llama_data_write_dummy data_ctx;
-    return llama_state_seq_get_data_internal(ctx, data_ctx, seq_id);
-}
+        // scheduler and compute buffers
+        {
+            // buffer types used for the compute buffer of each backend
+            std::vector backend_buft;
+            std::vector backend_ptrs;
+            for (auto & backend : ctx->backends) {
+                auto * buft = ggml_backend_get_default_buffer_type(backend.get());
+                auto backend_type = ggml_backend_dev_type(ggml_backend_get_device(backend.get()));
+                if (backend_type == GGML_BACKEND_DEVICE_TYPE_CPU && !model->devices.empty()) {
+                    // use the host buffer of the first device CPU for faster transfer of the intermediate state
+                    auto * dev = model->devices[0];
+                    auto * host_buft = ggml_backend_dev_host_buffer_type(dev);
+                    if (host_buft) {
+                        buft = host_buft;
+                    }
+                }
+                backend_buft.push_back(buft);
+                backend_ptrs.push_back(backend.get());
+            }
 
-size_t llama_state_seq_get_data(struct llama_context * ctx, uint8_t * dst, size_t size, llama_seq_id seq_id) {
-    llama_data_write_buffer data_ctx(dst, size);
-    try {
-        return llama_state_seq_get_data_internal(ctx, data_ctx, seq_id);
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: error saving sequence state: %s\n", __func__, err.what());
-        return 0;
-    }
-}
+            const size_t max_nodes = llama_model_max_nodes(*model);
 
-static size_t llama_state_seq_set_data_internal(struct llama_context * ctx, llama_data_read & data_ctx, llama_seq_id dest_seq_id) {
-    llama_synchronize(ctx);
+            // buffer used to store the computation graph and the tensor meta data
+            ctx->buf_compute_meta.resize(ggml_tensor_overhead()*max_nodes + ggml_graph_overhead_custom(max_nodes, false));
 
-    data_ctx.read_kv_cache(ctx, dest_seq_id);
+            // TODO: move these checks to ggml_backend_sched
+            // enabling pipeline parallelism in the scheduler increases memory usage, so it is only done when necessary
+            bool pipeline_parallel =
+                llama_get_device_count(*model) > 1 &&
+                model->n_gpu_layers > (int)model->hparams.n_layer &&
+                model->split_mode == LLAMA_SPLIT_MODE_LAYER &&
+                params.offload_kqv;
 
-    return data_ctx.get_size_read();
-}
+            // pipeline parallelism requires support for async compute and events in all devices
+            if (pipeline_parallel) {
+                for (auto & backend : ctx->backends) {
+                    auto dev_type = ggml_backend_dev_type(ggml_backend_get_device(backend.get()));
+                    if (dev_type == GGML_BACKEND_DEVICE_TYPE_CPU) {
+                        // ignore CPU backend
+                        continue;
+                    }
+                    auto * dev = ggml_backend_get_device(backend.get());
+                    ggml_backend_dev_props props;
+                    ggml_backend_dev_get_props(dev, &props);
+                    if (!props.caps.async || !props.caps.events) {
+                        // device does not support async compute or events
+                        pipeline_parallel = false;
+                        break;
+                    }
+                }
+            }
 
-size_t llama_state_seq_set_data(struct llama_context * ctx, const uint8_t * src, size_t size, llama_seq_id dest_seq_id) {
-    llama_data_read_buffer data_ctx(src, size);
-    try {
-        return llama_state_seq_set_data_internal(ctx, data_ctx, dest_seq_id);
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: error loading sequence state: %s\n", __func__, err.what());
-        return 0;
-    }
-}
+            ctx->sched.reset(ggml_backend_sched_new(backend_ptrs.data(), backend_buft.data(), backend_ptrs.size(), max_nodes, pipeline_parallel));
 
-static size_t llama_state_seq_save_file_internal(struct llama_context * ctx, const char * filepath, llama_seq_id seq_id, const llama_token * tokens, size_t n_token_count) {
-    llama_file file(filepath, "wb");
+            if (pipeline_parallel) {
+                LLAMA_LOG_INFO("%s: pipeline parallelism enabled (n_copies=%d)\n", __func__, ggml_backend_sched_get_n_copies(ctx->sched.get()));
+            }
 
-    file.write_u32(LLAMA_STATE_SEQ_MAGIC);
-    file.write_u32(LLAMA_STATE_SEQ_VERSION);
+            // initialize scheduler with the worst-case graph
+            uint32_t n_seqs = 1; // TODO: worst-case number of sequences
+            uint32_t n_tokens = std::min(cparams.n_ctx, cparams.n_ubatch);
+            llama_token token = llama_token_bos(&ctx->model); // not actually used by llama_build_graph, but required to choose between token and embedding inputs graph
 
-    // save the prompt
-    file.write_u32((uint32_t) n_token_count);
-    file.write_raw(tokens, sizeof(llama_token) * n_token_count);
+            llama_ubatch ubatch_pp = { true, n_tokens, n_tokens / n_seqs, n_seqs, &token, nullptr, nullptr, nullptr, nullptr, nullptr};
+            ggml_cgraph * gf_pp = llama_build_graph(*ctx, ubatch_pp, true);
 
-    // save the context state using stream saving
-    llama_data_write_file data_ctx(&file);
-    llama_state_seq_get_data_internal(ctx, data_ctx, seq_id);
+            // reserve pp graph first so that buffers are only allocated once
+            ggml_backend_sched_reserve(ctx->sched.get(), gf_pp);
+            int n_splits_pp = ggml_backend_sched_get_n_splits(ctx->sched.get());
+            int n_nodes_pp = ggml_graph_n_nodes(gf_pp);
 
-    const size_t res = file.tell();
-    GGML_ASSERT(res == sizeof(uint32_t) * 3 + sizeof(llama_token) * n_token_count + data_ctx.get_size_written());
-    return res;
-}
+            // reserve with tg graph to get the number of splits and nodes
+            llama_ubatch ubatch_tg = { true, 1, 1, n_seqs, &token, nullptr, nullptr, nullptr, nullptr, nullptr};
+            ggml_cgraph * gf_tg = llama_build_graph(*ctx, ubatch_tg, true);
+            ggml_backend_sched_reserve(ctx->sched.get(), gf_tg);
+            int n_splits_tg = ggml_backend_sched_get_n_splits(ctx->sched.get());
+            int n_nodes_tg = ggml_graph_n_nodes(gf_tg);
 
-static size_t llama_state_seq_load_file_internal(struct llama_context * ctx, const char * filepath, llama_seq_id dest_seq_id, llama_token * tokens_out, size_t n_token_capacity, size_t * n_token_count_out) {
-    llama_file file(filepath, "rb");
+            // reserve again with pp graph to avoid ggml-alloc reallocations during inference
+            gf_pp = llama_build_graph(*ctx, ubatch_pp, true);
+            if (!ggml_backend_sched_reserve(ctx->sched.get(), gf_pp)) {
+                LLAMA_LOG_ERROR("%s: failed to allocate compute buffers\n", __func__);
+                llama_free(ctx);
+                return nullptr;
+            }
 
-    // version checks
-    {
-        const uint32_t magic   = file.read_u32();
-        const uint32_t version = file.read_u32();
+            for (size_t i = 0; i < backend_ptrs.size(); ++i) {
+                ggml_backend_t backend = backend_ptrs[i];
+                ggml_backend_buffer_type_t buft = backend_buft[i];
+                size_t size = ggml_backend_sched_get_buffer_size(ctx->sched.get(), backend);
+                if (size > 1) {
+                    LLAMA_LOG_INFO("%s: %10s compute buffer size = %8.2f MiB\n", __func__,
+                            ggml_backend_buft_name(buft),
+                            size / 1024.0 / 1024.0);
+                }
+            }
 
-        if (magic != LLAMA_STATE_SEQ_MAGIC || version != LLAMA_STATE_SEQ_VERSION) {
-            LLAMA_LOG_ERROR("%s: unknown (magic, version) for sequence state file: %08x, %08x\n", __func__, magic, version);
-            return 0;
+            if (n_nodes_pp == n_nodes_tg) {
+                LLAMA_LOG_INFO("%s: graph nodes  = %d\n", __func__, n_nodes_pp);
+            } else {
+                LLAMA_LOG_INFO("%s: graph nodes  = %d (with bs=%d), %d (with bs=1)\n", __func__, n_nodes_pp, n_tokens, n_nodes_tg);
+            }
+            if (n_splits_pp == n_splits_tg) {
+                LLAMA_LOG_INFO("%s: graph splits = %d\n", __func__, n_splits_pp);
+            } else {
+                LLAMA_LOG_INFO("%s: graph splits = %d (with bs=%d), %d (with bs=1)\n", __func__, n_splits_pp, n_tokens, n_splits_tg);
+            }
         }
     }
 
-    // load the prompt
-    {
-        const uint32_t n_token_count = file.read_u32();
-
-        if (n_token_count > n_token_capacity) {
-            LLAMA_LOG_ERROR("%s: token count in sequence state file exceeded capacity! %u > %zu\n", __func__, n_token_count, n_token_capacity);
-            return 0;
-        }
+    return ctx;
+}
 
-        file.read_raw(tokens_out, sizeof(llama_token) * n_token_count);
-        *n_token_count_out = n_token_count;
-    }
+//
+// kv cache
+//
 
-    // restore the context state
-    {
-        const size_t state_size = file.size - file.tell();
-        llama_data_read_file data_ctx(&file);
-        const size_t nread = llama_state_seq_set_data_internal(ctx, data_ctx, dest_seq_id);
-        if (!nread) {
-            LLAMA_LOG_ERROR("%s: failed to restore sequence state\n", __func__);
-            return 0;
-        }
-        GGML_ASSERT(nread <= state_size);
-        GGML_ASSERT(nread + sizeof(uint32_t) * 3 + sizeof(llama_token) * *n_token_count_out == file.tell());
-    }
+// TODO: tmp bridges below until `struct llama_kv_cache` is exposed through the public API
 
-    return file.tell();
+struct llama_kv_cache_view llama_kv_cache_view_init(const struct llama_context * ctx, int32_t n_seq_max) {
+    return llama_kv_cache_view_init(ctx->kv_self, n_seq_max);
 }
 
-size_t llama_state_seq_save_file(struct llama_context * ctx, const char * filepath, llama_seq_id seq_id, const llama_token * tokens, size_t n_token_count) {
-    try {
-        return llama_state_seq_save_file_internal(ctx, filepath, seq_id, tokens, n_token_count);
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: error saving sequence state file: %s\n", __func__, err.what());
-        return 0;
-    }
+void llama_kv_cache_view_update(const struct llama_context * ctx, struct llama_kv_cache_view * view) {
+    llama_kv_cache_view_update(view, ctx->kv_self);
 }
 
-size_t llama_state_seq_load_file(struct llama_context * ctx, const char * filepath, llama_seq_id dest_seq_id, llama_token * tokens_out, size_t n_token_capacity, size_t * n_token_count_out) {
-    try {
-        return llama_state_seq_load_file_internal(ctx, filepath, dest_seq_id, tokens_out, n_token_capacity, n_token_count_out);
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: error loading sequence state file: %s\n", __func__, err.what());
-        return 0;
-    }
+int32_t llama_get_kv_cache_token_count(const struct llama_context * ctx) {
+    return llama_get_kv_cache_token_count(ctx->kv_self);
 }
 
-void llama_set_n_threads(struct llama_context * ctx, int32_t n_threads, int32_t n_threads_batch) {
-    ctx->cparams.n_threads       = n_threads;
-    ctx->cparams.n_threads_batch = n_threads_batch;
+int32_t llama_get_kv_cache_used_cells(const struct llama_context * ctx) {
+    return llama_get_kv_cache_used_cells(ctx->kv_self);
 }
 
-int32_t llama_n_threads(struct llama_context * ctx) {
-    return ctx->cparams.n_threads;
+void llama_kv_cache_clear(struct llama_context * ctx) {
+    llama_kv_cache_clear(ctx->kv_self);
 }
 
-int32_t llama_n_threads_batch(struct llama_context * ctx) {
-    return ctx->cparams.n_threads_batch;
+bool llama_kv_cache_seq_rm(struct llama_context * ctx, llama_seq_id seq_id, llama_pos p0, llama_pos p1) {
+    return llama_kv_cache_seq_rm(ctx->kv_self, seq_id, p0, p1);
 }
 
-void llama_set_abort_callback(struct llama_context * ctx, bool (*abort_callback)(void * data), void * abort_callback_data) {
-    ctx->abort_callback      = abort_callback;
-    ctx->abort_callback_data = abort_callback_data;
-
-    for (auto & backend : ctx->backends) {
-        auto * reg = ggml_backend_dev_backend_reg(ggml_backend_get_device(backend.get()));
-        auto * set_abort_callback_fn = (ggml_backend_set_abort_callback_t) ggml_backend_reg_get_proc_address(reg, "ggml_backend_set_abort_callback");
-        if (set_abort_callback_fn) {
-            set_abort_callback_fn(backend.get(), ctx->abort_callback, ctx->abort_callback_data);
-        }
+void llama_kv_cache_seq_cp(struct llama_context * ctx, llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) {
+    if (seq_id_src == seq_id_dst) {
+        return;
     }
+    llama_kv_cache_seq_cp(ctx->kv_self, seq_id_src, seq_id_dst, p0, p1);
 }
 
-void llama_set_embeddings(struct llama_context * ctx, bool embeddings) {
-    ctx->cparams.embeddings = embeddings;
+void llama_kv_cache_seq_keep(struct llama_context * ctx, llama_seq_id seq_id) {
+    llama_kv_cache_seq_keep(ctx->kv_self, seq_id);
 }
 
-void llama_set_causal_attn(struct llama_context * ctx, bool causal_attn) {
-    ctx->cparams.causal_attn = causal_attn;
-}
+void llama_kv_cache_seq_add(struct llama_context * ctx, llama_seq_id seq_id, llama_pos p0, llama_pos p1, llama_pos delta) {
+    if (delta == 0) {
+        return;
+    }
 
-struct llama_batch llama_batch_get_one(
-             llama_token * tokens,
-                 int32_t   n_tokens) {
-    return {
-        /*n_tokens       =*/ n_tokens,
-        /*tokens         =*/ tokens,
-        /*embd           =*/ nullptr,
-        /*pos            =*/ nullptr,
-        /*n_seq_id       =*/ nullptr,
-        /*seq_id         =*/ nullptr,
-        /*logits         =*/ nullptr,
-    };
+    llama_kv_cache_seq_add(ctx->kv_self, seq_id, p0, p1, delta);
 }
 
-struct llama_batch llama_batch_init(int32_t n_tokens_alloc, int32_t embd, int32_t n_seq_max) {
-    llama_batch batch = {
-        /*n_tokens       =*/ 0,
-        /*tokens         =*/ nullptr,
-        /*embd           =*/ nullptr,
-        /*pos            =*/ nullptr,
-        /*n_seq_id       =*/ nullptr,
-        /*seq_id         =*/ nullptr,
-        /*logits         =*/ nullptr,
-    };
-
-    if (embd) {
-        batch.embd = (float *) malloc(sizeof(float) * n_tokens_alloc * embd);
-    } else {
-        batch.token = (llama_token *) malloc(sizeof(llama_token) * n_tokens_alloc);
+void llama_kv_cache_seq_div(struct llama_context * ctx, llama_seq_id seq_id, llama_pos p0, llama_pos p1, int d) {
+    if (d == 1) {
+        return;
     }
 
-    batch.pos      = (llama_pos *)     malloc(sizeof(llama_pos)      * n_tokens_alloc);
-    batch.n_seq_id = (int32_t *)       malloc(sizeof(int32_t)        * n_tokens_alloc);
-    batch.seq_id   = (llama_seq_id **) malloc(sizeof(llama_seq_id *) * (n_tokens_alloc + 1));
-    for (int i = 0; i < n_tokens_alloc; ++i) {
-        batch.seq_id[i] = (llama_seq_id *) malloc(sizeof(llama_seq_id) * n_seq_max);
-    }
-    batch.seq_id[n_tokens_alloc] = nullptr;
+    llama_kv_cache_seq_div(ctx->kv_self, seq_id, p0, p1, d);
+}
 
-    batch.logits   = (int8_t *)        malloc(sizeof(int8_t)         * n_tokens_alloc);
+llama_pos llama_kv_cache_seq_pos_max(struct llama_context * ctx, llama_seq_id seq_id) {
+    return llama_kv_cache_seq_pos_max(ctx->kv_self, seq_id);
+}
 
-    return batch;
+void llama_kv_cache_defrag(struct llama_context * ctx) {
+    llama_kv_cache_defrag(ctx->kv_self);
 }
 
-void llama_batch_free(struct llama_batch batch) {
-    if (batch.token)    free(batch.token);
-    if (batch.embd)     free(batch.embd);
-    if (batch.pos)      free(batch.pos);
-    if (batch.n_seq_id) free(batch.n_seq_id);
-    if (batch.seq_id) {
-        for (int i = 0; batch.seq_id[i] != nullptr; ++i) {
-            free(batch.seq_id[i]);
-        }
-        free(batch.seq_id);
-    }
-    if (batch.logits)   free(batch.logits);
+void llama_kv_cache_update(struct llama_context * ctx) {
+    llama_kv_cache_update_impl(*ctx);
+}
+
+bool llama_kv_cache_can_shift(struct llama_context * ctx) {
+    return llama_kv_cache_can_shift(ctx->kv_self);
 }
 
+///
+
 int32_t llama_encode(
         struct llama_context * ctx,
           struct llama_batch   batch) {
-    const int ret = llama_encode_internal(*ctx, batch);
+    const int ret = llama_encode_impl(*ctx, batch);
     if (ret != 0) {
         LLAMA_LOG_ERROR("%s: failed to encode, ret = %d\n", __func__, ret);
     }
@@ -21896,7 +12214,7 @@ int32_t llama_encode(
 int32_t llama_decode(
         struct llama_context * ctx,
           struct llama_batch   batch) {
-    const int ret = llama_decode_internal(*ctx, batch);
+    const int ret = llama_decode_impl(*ctx, batch);
     if (ret != 0) {
         LLAMA_LOG_ERROR("%s: failed to decode, ret = %d\n", __func__, ret);
     }
@@ -21904,150 +12222,12 @@ int32_t llama_decode(
     return ret;
 }
 
-void llama_synchronize(struct llama_context * ctx) {
-    ggml_backend_sched_synchronize(ctx->sched.get());
-
-    // FIXME: if multiple single tokens are evaluated without a synchronization,
-    // the stats will be added to the prompt evaluation stats
-    // this should only happen when using batch size 1 to evaluate a batch
-
-    // add the evaluation to the stats
-    if (ctx->n_queued_tokens == 1) {
-        if (!ctx->cparams.no_perf) {
-            ctx->t_eval_us += ggml_time_us() - ctx->t_compute_start_us;
-        }
-        ctx->n_eval++;
-    } else if (ctx->n_queued_tokens > 1) {
-        if (!ctx->cparams.no_perf) {
-            ctx->t_p_eval_us += ggml_time_us() - ctx->t_compute_start_us;
-        }
-        ctx->n_p_eval += ctx->n_queued_tokens;
-    }
-
-    // get a more accurate load time, upon first eval
-    if (ctx->n_queued_tokens > 0 && !ctx->has_evaluated_once) {
-        ctx->t_load_us = ggml_time_us() - ctx->t_start_us;
-        ctx->has_evaluated_once = true;
-    }
-
-    ctx->n_queued_tokens = 0;
-    ctx->t_compute_start_us = 0;
-}
-
-float * llama_get_logits(struct llama_context * ctx) {
-    llama_synchronize(ctx);
-
-    // reorder logits for backward compatibility
-    // TODO: maybe deprecate this
-    llama_output_reorder(ctx);
-
-    return ctx->logits;
-}
-
-float * llama_get_logits_ith(struct llama_context * ctx, int32_t i) {
-    int32_t j = -1;
-    llama_synchronize(ctx);
-
-    try {
-        if (ctx->logits == nullptr) {
-            throw std::runtime_error("no logits");
-        }
-
-        if (i < 0) {
-            j = ctx->n_outputs + i;
-            if (j < 0) {
-                throw std::runtime_error(format("negative index out of range [0, %d)", ctx->n_outputs));
-            }
-        } else if ((size_t) i >= ctx->output_ids.size()) {
-            throw std::runtime_error(format("out of range [0, %zu)", ctx->output_ids.size()));
-        } else {
-            j = ctx->output_ids[i];
-        }
-
-        if (j < 0) {
-            throw std::runtime_error(format("batch.logits[%d] != true", i));
-        }
-        if (j >= ctx->n_outputs) {
-            // This should not happen
-            throw std::runtime_error(format("corrupt output buffer (j=%d, n_outputs=%d)", j, ctx->n_outputs));
-        }
-
-        return ctx->logits + j*ctx->model.hparams.n_vocab;
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: invalid logits id %d, reason: %s\n", __func__, i, err.what());
-#ifndef NDEBUG
-        GGML_ABORT("fatal error");
-#else
-        return nullptr;
-#endif
-    }
-}
-
-float * llama_get_embeddings(struct llama_context * ctx) {
-    llama_synchronize(ctx);
-
-    // reorder embeddings for backward compatibility
-    // TODO: maybe deprecate this
-    llama_output_reorder(ctx);
-
-    return ctx->embd;
-}
-
-float * llama_get_embeddings_ith(struct llama_context * ctx, int32_t i) {
-    int32_t j = -1;
-
-    llama_synchronize(ctx);
-
-    try {
-        if (ctx->embd == nullptr) {
-            throw std::runtime_error("no embeddings");
-        }
-
-        if (i < 0) {
-            j = ctx->n_outputs + i;
-            if (j < 0) {
-                throw std::runtime_error(format("negative index out of range [0, %d)", ctx->n_outputs));
-            }
-        } else if ((size_t) i >= ctx->output_ids.size()) {
-            throw std::runtime_error(format("out of range [0, %zu)", ctx->output_ids.size()));
-        } else {
-            j = ctx->output_ids[i];
-        }
-
-        if (j < 0) {
-            throw std::runtime_error(format("batch.logits[%d] != true", i));
-        }
-        if (j >= ctx->n_outputs) {
-            // This should not happen
-            throw std::runtime_error(format("corrupt output buffer (j=%d, n_outputs=%d)", j, ctx->n_outputs));
-        }
-
-        return ctx->embd + j*ctx->model.hparams.n_embd;
-    } catch (const std::exception & err) {
-        LLAMA_LOG_ERROR("%s: invalid embeddings id %d, reason: %s\n", __func__, i, err.what());
-#ifndef NDEBUG
-        GGML_ABORT("fatal error");
-#else
-        return nullptr;
-#endif
-    }
-}
-
-float * llama_get_embeddings_seq(struct llama_context * ctx, llama_seq_id seq_id) {
-    llama_synchronize(ctx);
-
-    auto it = ctx->embd_seq.find(seq_id);
-    if (it == ctx->embd_seq.end()) {
-        return nullptr;
-    }
-
-    return it->second.data();
-}
-
 //
 // vocab
 //
 
+// TODO: tmp bridges below until `struct llama_vocab` is exposed through the public API
+
 const char * llama_token_get_text(const struct llama_model * model, llama_token token) {
     return llama_token_get_text_impl(model->vocab, token);
 }
@@ -22180,466 +12360,6 @@ int32_t llama_detokenize(
 // chat templates
 //
 
-static llm_chat_template llama_chat_detect_template(const std::string & tmpl) {
-    if (LLM_CHAT_TEMPLATES.find(tmpl) != LLM_CHAT_TEMPLATES.end()) {
-        return LLM_CHAT_TEMPLATES.at(tmpl);
-    }
-    auto tmpl_contains = [&tmpl](const char * haystack) -> bool {
-        return tmpl.find(haystack) != std::string::npos;
-    };
-    if (tmpl_contains("<|im_start|>")) {
-        return LLM_CHAT_TEMPLATE_CHATML;
-    } else if (tmpl.find("mistral") == 0 || tmpl_contains("[INST]")) {
-        if (tmpl_contains("[SYSTEM_PROMPT]")) {
-            return LLM_CHAT_TEMPLATE_MISTRAL_V7;
-        } else if (
-            // catches official 'v1' template
-            tmpl_contains("' [INST] ' + system_message")
-            // catches official 'v3' and 'v3-tekken' templates
-            || tmpl_contains("[AVAILABLE_TOOLS]")
-        ) {
-            // Official mistral 'v1', 'v3' and 'v3-tekken' templates
-            // See: https://github.com/mistralai/cookbook/blob/main/concept-deep-dive/tokenization/chat_templates.md
-            // See: https://github.com/mistralai/cookbook/blob/main/concept-deep-dive/tokenization/templates.md
-            if (tmpl_contains(" [INST]")) {
-                return LLM_CHAT_TEMPLATE_MISTRAL_V1;
-            } else if (tmpl_contains("\"[INST]\"")) {
-                return LLM_CHAT_TEMPLATE_MISTRAL_V3_TEKKEN;
-            }
-            return LLM_CHAT_TEMPLATE_MISTRAL_V3;
-        } else {
-            // llama2 template and its variants
-            // [variant] support system message
-            // See: https://huggingface.co/blog/llama2#how-to-prompt-llama-2
-            bool support_system_message = tmpl_contains("<>");
-            bool add_bos_inside_history = tmpl_contains("bos_token + '[INST]");
-            bool strip_message = tmpl_contains("content.strip()");
-            if (strip_message) {
-                return LLM_CHAT_TEMPLATE_LLAMA_2_SYS_STRIP;
-            } else if (add_bos_inside_history) {
-                return LLM_CHAT_TEMPLATE_LLAMA_2_SYS_BOS;
-            } else if (support_system_message) {
-                return LLM_CHAT_TEMPLATE_LLAMA_2_SYS;
-            } else {
-                return LLM_CHAT_TEMPLATE_LLAMA_2;
-            }
-        }
-    } else if (tmpl_contains("<|assistant|>") && tmpl_contains("<|end|>")) {
-        return LLM_CHAT_TEMPLATE_PHI_3;
-    } else if (tmpl_contains("<|assistant|>") && tmpl_contains("<|user|>")) {
-        return LLM_CHAT_TEMPLATE_FALCON_3;
-    } else if (tmpl_contains("<|user|>") && tmpl_contains("<|endoftext|>")) {
-        return LLM_CHAT_TEMPLATE_ZEPHYR;
-    } else if (tmpl_contains("bos_token + message['role']")) {
-        return LLM_CHAT_TEMPLATE_MONARCH;
-    } else if (tmpl_contains("")) {
-        return LLM_CHAT_TEMPLATE_GEMMA;
-    } else if (tmpl_contains("'\\n\\nAssistant: ' + eos_token")) {
-        // OrionStarAI/Orion-14B-Chat
-        return LLM_CHAT_TEMPLATE_ORION;
-    } else if (tmpl_contains("GPT4 Correct ")) {
-        // openchat/openchat-3.5-0106
-        return LLM_CHAT_TEMPLATE_OPENCHAT;
-    } else if (tmpl_contains("USER: ") && tmpl_contains("ASSISTANT: ")) {
-        // eachadea/vicuna-13b-1.1 (and Orca variant)
-        if (tmpl_contains("SYSTEM: ")) {
-            return LLM_CHAT_TEMPLATE_VICUNA_ORCA;
-        }
-        return LLM_CHAT_TEMPLATE_VICUNA;
-    } else if (tmpl_contains("### Instruction:") && tmpl_contains("<|EOT|>")) {
-        // deepseek-ai/deepseek-coder-33b-instruct
-        return LLM_CHAT_TEMPLATE_DEEPSEEK;
-    } else if (tmpl_contains("<|START_OF_TURN_TOKEN|>") && tmpl_contains("<|USER_TOKEN|>")) {
-        // CohereForAI/c4ai-command-r-plus
-        return LLM_CHAT_TEMPLATE_COMMAND_R;
-    } else if (tmpl_contains("<|start_header_id|>") && tmpl_contains("<|end_header_id|>")) {
-        return LLM_CHAT_TEMPLATE_LLAMA_3;
-    } else if (tmpl_contains("[gMASK]sop")) {
-        // chatglm3-6b
-        return LLM_CHAT_TEMPLATE_CHATGML_3;
-    } else if (tmpl_contains("[gMASK]")) {
-        return LLM_CHAT_TEMPLATE_CHATGML_4;
-    } else if (tmpl_contains(LU8("<用户>"))) {
-        // MiniCPM-3B-OpenHermes-2.5-v2-GGUF
-        return LLM_CHAT_TEMPLATE_MINICPM;
-    } else if (tmpl_contains("'Assistant: ' + message['content'] + eos_token")) {
-        return LLM_CHAT_TEMPLATE_DEEPSEEK_2;
-    } else if (tmpl_contains("[|system|]") && tmpl_contains("[|assistant|]") && tmpl_contains("[|endofturn|]")) {
-        // ref: https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct/discussions/8#66bae61b1893d14ee8ed85bb
-        // EXAONE-3.0-7.8B-Instruct
-        return LLM_CHAT_TEMPLATE_EXAONE_3;
-    } else if (tmpl_contains("rwkv-world")) {
-        return LLM_CHAT_TEMPLATE_RWKV_WORLD;
-    } else if (tmpl_contains("<|start_of_role|>")) {
-        return LLM_CHAT_TEMPLATE_GRANITE;
-    } else if (tmpl_contains("message['role'] + additional_special_tokens[0] + message['content'] + additional_special_tokens[1]")) {
-        return LLM_CHAT_TEMPLATE_GIGACHAT;
-    }
-    return LLM_CHAT_TEMPLATE_UNKNOWN;
-}
-
-// Simple version of "llama_apply_chat_template" that only works with strings
-// This function uses heuristic checks to determine commonly used template. It is not a jinja parser.
-static int32_t llama_chat_apply_template_internal(
-    const llm_chat_template tmpl,
-    const std::vector & chat,
-    std::string & dest, bool add_ass) {
-    // Taken from the research: https://github.com/ggerganov/llama.cpp/issues/5527
-    std::stringstream ss;
-    if (tmpl == LLM_CHAT_TEMPLATE_CHATML) {
-        // chatml template
-        for (auto message : chat) {
-            ss << "<|im_start|>" << message->role << "\n" << message->content << "<|im_end|>\n";
-        }
-        if (add_ass) {
-            ss << "<|im_start|>assistant\n";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_MISTRAL_V7) {
-        // Official mistral 'v7' template
-        // See: https://huggingface.co/mistralai/Mistral-Large-Instruct-2411#basic-instruct-template-v7
-        for (auto message : chat) {
-            std::string role(message->role);
-            std::string content(message->content);
-            if (role == "system") {
-                ss << "[SYSTEM_PROMPT] " << content << "[/SYSTEM_PROMPT]";
-            } else if (role == "user") {
-                ss << "[INST] " << content << "[/INST]";
-            }
-            else {
-                ss << " " << content << "";
-            }
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_MISTRAL_V1
-            || tmpl == LLM_CHAT_TEMPLATE_MISTRAL_V3
-            || tmpl == LLM_CHAT_TEMPLATE_MISTRAL_V3_TEKKEN) {
-        // See: https://github.com/mistralai/cookbook/blob/main/concept-deep-dive/tokenization/chat_templates.md
-        // See: https://github.com/mistralai/cookbook/blob/main/concept-deep-dive/tokenization/templates.md
-        std::string leading_space = tmpl == LLM_CHAT_TEMPLATE_MISTRAL_V1 ? " " : "";
-        std::string trailing_space = tmpl == LLM_CHAT_TEMPLATE_MISTRAL_V3_TEKKEN ? "" : " ";
-        bool trim_assistant_message = tmpl == LLM_CHAT_TEMPLATE_MISTRAL_V3;
-        bool is_inside_turn = false;
-        for (auto message : chat) {
-            if (!is_inside_turn) {
-                ss << leading_space << "[INST]" << trailing_space;
-                is_inside_turn = true;
-            }
-            std::string role(message->role);
-            std::string content(message->content);
-            if (role == "system") {
-                ss << content << "\n\n";
-            } else if (role == "user") {
-                ss << content << leading_space << "[/INST]";
-            } else {
-                ss << trailing_space << (trim_assistant_message ? trim(content) : content) << "";
-                is_inside_turn = false;
-            }
-        }
-    } else if (
-            tmpl == LLM_CHAT_TEMPLATE_LLAMA_2
-            || tmpl == LLM_CHAT_TEMPLATE_LLAMA_2_SYS
-            || tmpl == LLM_CHAT_TEMPLATE_LLAMA_2_SYS_BOS
-            || tmpl == LLM_CHAT_TEMPLATE_LLAMA_2_SYS_STRIP) {
-        // llama2 template and its variants
-        // [variant] support system message
-        // See: https://huggingface.co/blog/llama2#how-to-prompt-llama-2
-        bool support_system_message = tmpl != LLM_CHAT_TEMPLATE_LLAMA_2;
-        // [variant] add BOS inside history
-        bool add_bos_inside_history = tmpl == LLM_CHAT_TEMPLATE_LLAMA_2_SYS_BOS;
-        // [variant] trim spaces from the input message
-        bool strip_message = tmpl == LLM_CHAT_TEMPLATE_LLAMA_2_SYS_STRIP;
-        // construct the prompt
-        bool is_inside_turn = true; // skip BOS at the beginning
-        ss << "[INST] ";
-        for (auto message : chat) {
-            std::string content = strip_message ? trim(message->content) : message->content;
-            std::string role(message->role);
-            if (!is_inside_turn) {
-                is_inside_turn = true;
-                ss << (add_bos_inside_history ? "[INST] " : "[INST] ");
-            }
-            if (role == "system") {
-                if (support_system_message) {
-                    ss << "<>\n" << content << "\n<>\n\n";
-                } else {
-                    // if the model does not support system message, we still include it in the first message, but without <>
-                    ss << content << "\n";
-                }
-            } else if (role == "user") {
-                ss << content << " [/INST]";
-            } else {
-                ss << content << "";
-                is_inside_turn = false;
-            }
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_PHI_3) {
-        // Phi 3
-        for (auto message : chat) {
-            std::string role(message->role);
-            ss << "<|" << role << "|>\n" << message->content << "<|end|>\n";
-        }
-        if (add_ass) {
-            ss << "<|assistant|>\n";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_FALCON_3) {
-        // Falcon 3
-        for (auto message : chat) {
-            std::string role(message->role);
-            ss << "<|" << role << "|>\n" << message->content << "\n";
-        }
-        if (add_ass) {
-            ss << "<|assistant|>\n";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_ZEPHYR) {
-        // zephyr template
-        for (auto message : chat) {
-            ss << "<|" << message->role << "|>" << "\n" << message->content << "<|endoftext|>\n";
-        }
-        if (add_ass) {
-            ss << "<|assistant|>\n";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_MONARCH) {
-        // mlabonne/AlphaMonarch-7B template (the  is included inside history)
-        for (auto message : chat) {
-            std::string bos = (message == chat.front()) ? "" : ""; // skip BOS for first message
-            ss << bos << message->role << "\n" << message->content << "\n";
-        }
-        if (add_ass) {
-            ss << "assistant\n";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_GEMMA) {
-        // google/gemma-7b-it
-        std::string system_prompt = "";
-        for (auto message : chat) {
-            std::string role(message->role);
-            if (role == "system") {
-                // there is no system message for gemma, but we will merge it with user prompt, so nothing is broken
-                system_prompt = trim(message->content);
-                continue;
-            }
-            // in gemma, "assistant" is "model"
-            role = role == "assistant" ? "model" : message->role;
-            ss << "" << role << "\n";
-            if (!system_prompt.empty() && role != "model") {
-                ss << system_prompt << "\n\n";
-                system_prompt = "";
-            }
-            ss << trim(message->content) << "\n";
-        }
-        if (add_ass) {
-            ss << "model\n";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_ORION) {
-        // OrionStarAI/Orion-14B-Chat
-        std::string system_prompt = "";
-        for (auto message : chat) {
-            std::string role(message->role);
-            if (role == "system") {
-                // there is no system message support, we will merge it with user prompt
-                system_prompt = message->content;
-                continue;
-            } else if (role == "user") {
-                ss << "Human: ";
-                if (!system_prompt.empty()) {
-                    ss << system_prompt << "\n\n";
-                    system_prompt = "";
-                }
-                ss << message->content << "\n\nAssistant: ";
-            } else {
-                ss << message->content << "";
-            }
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_OPENCHAT) {
-        // openchat/openchat-3.5-0106,
-        for (auto message : chat) {
-            std::string role(message->role);
-            if (role == "system") {
-                ss << message->content << "<|end_of_turn|>";
-            } else {
-                role[0] = toupper(role[0]);
-                ss << "GPT4 Correct " << role << ": " << message->content << "<|end_of_turn|>";
-            }
-        }
-        if (add_ass) {
-            ss << "GPT4 Correct Assistant:";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_VICUNA || tmpl == LLM_CHAT_TEMPLATE_VICUNA_ORCA) {
-        // eachadea/vicuna-13b-1.1 (and Orca variant)
-        for (auto message : chat) {
-            std::string role(message->role);
-            if (role == "system") {
-                // Orca-Vicuna variant uses a system prefix
-                if (tmpl == LLM_CHAT_TEMPLATE_VICUNA_ORCA) {
-                    ss << "SYSTEM: " << message->content << "\n";
-                } else {
-                    ss << message->content << "\n\n";
-                }
-            } else if (role == "user") {
-                ss << "USER: " << message->content << "\n";
-            } else if (role == "assistant") {
-                ss << "ASSISTANT: " << message->content << "\n";
-            }
-        }
-        if (add_ass) {
-            ss << "ASSISTANT:";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_DEEPSEEK) {
-        // deepseek-ai/deepseek-coder-33b-instruct
-        for (auto message : chat) {
-            std::string role(message->role);
-            if (role == "system") {
-                ss << message->content;
-            } else if (role == "user") {
-                ss << "### Instruction:\n" << message->content << "\n";
-            } else if (role == "assistant") {
-                ss << "### Response:\n" << message->content << "\n<|EOT|>\n";
-            }
-        }
-        if (add_ass) {
-            ss << "### Response:\n";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_COMMAND_R) {
-        // CohereForAI/c4ai-command-r-plus
-        for (auto message : chat) {
-            std::string role(message->role);
-            if (role == "system") {
-                ss << "<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>" << trim(message->content) << "<|END_OF_TURN_TOKEN|>";
-            } else if (role == "user") {
-                ss << "<|START_OF_TURN_TOKEN|><|USER_TOKEN|>" << trim(message->content) << "<|END_OF_TURN_TOKEN|>";
-            } else if (role == "assistant") {
-                ss << "<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>" << trim(message->content) << "<|END_OF_TURN_TOKEN|>";
-            }
-        }
-        if (add_ass) {
-            ss << "<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_LLAMA_3) {
-        // Llama 3
-        for (auto message : chat) {
-            std::string role(message->role);
-            ss << "<|start_header_id|>" << role << "<|end_header_id|>\n\n" << trim(message->content) << "<|eot_id|>";
-        }
-        if (add_ass) {
-            ss << "<|start_header_id|>assistant<|end_header_id|>\n\n";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_CHATGML_3) {
-        // chatglm3-6b
-        ss << "[gMASK]" << "sop";
-        for (auto message : chat) {
-            std::string role(message->role);
-            ss << "<|" << role << "|>" << "\n " << message->content;
-        }
-        if (add_ass) {
-            ss << "<|assistant|>";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_CHATGML_4) {
-        ss << "[gMASK]" << "";
-        for (auto message : chat) {
-            std::string role(message->role);
-            ss << "<|" << role << "|>" << "\n" << message->content;
-        }
-        if (add_ass) {
-            ss << "<|assistant|>";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_MINICPM) {
-        // MiniCPM-3B-OpenHermes-2.5-v2-GGUF
-        for (auto message : chat) {
-            std::string role(message->role);
-            if (role == "user") {
-                ss << LU8("<用户>");
-                ss << trim(message->content);
-                ss << "";
-            } else {
-                ss << trim(message->content);
-            }
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_DEEPSEEK_2) {
-        // DeepSeek-V2
-        for (auto message : chat) {
-            std::string role(message->role);
-            if (role == "system") {
-                ss << message->content << "\n\n";
-            } else if (role == "user") {
-                ss << "User: " << message->content << "\n\n";
-            } else if (role == "assistant") {
-                ss << "Assistant: " << message->content << LU8("<｜end▁of▁sentence｜>");
-            }
-        }
-        if (add_ass) {
-            ss << "Assistant:";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_EXAONE_3) {
-        // ref: https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct/discussions/8#66bae61b1893d14ee8ed85bb
-        // EXAONE-3.0-7.8B-Instruct
-        for (auto message : chat) {
-            std::string role(message->role);
-            if (role == "system") {
-                ss << "[|system|]" << trim(message->content) << "[|endofturn|]\n";
-            } else if (role == "user") {
-                ss << "[|user|]" << trim(message->content) << "\n";
-            } else if (role == "assistant") {
-                ss << "[|assistant|]" << trim(message->content) << "[|endofturn|]\n";
-            }
-        }
-        if (add_ass) {
-            ss << "[|assistant|]";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_RWKV_WORLD) {
-        // this template requires the model to have "\n\n" as EOT token
-        for (auto message : chat) {
-            std::string role(message->role);
-            if (role == "user") {
-                ss << "User: " << message->content << "\n\nAssistant:";
-            } else {
-                ss << message->content << "\n\n";
-            }
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_GRANITE) {
-        // IBM Granite template
-        for (const auto & message : chat) {
-            std::string role(message->role);
-            ss << "<|start_of_role|>" << role << "<|end_of_role|>";
-            if (role == "assistant_tool_call") {
-                ss << "<|tool_call|>";
-            }
-            ss << message->content << "<|end_of_text|>\n";
-        }
-        if (add_ass) {
-            ss << "<|start_of_role|>assistant<|end_of_role|>\n";
-        }
-    } else if (tmpl == LLM_CHAT_TEMPLATE_GIGACHAT) {
-        // GigaChat template
-        bool has_system = !chat.empty() && std::string(chat[0]->role) == "system";
-
-        // Handle system message if present
-        if (has_system) {
-            ss << "" << chat[0]->content << "<|message_sep|>";
-        } else {
-            ss << "";
-        }
-
-        // Process remaining messages
-        for (size_t i = has_system ? 1 : 0; i < chat.size(); i++) {
-            std::string role(chat[i]->role);
-            if (role == "user") {
-                ss << "user<|role_sep|>" << chat[i]->content << "<|message_sep|>"
-                << "available functions<|role_sep|>[]<|message_sep|>";
-            } else if (role == "assistant") {
-                ss << "assistant<|role_sep|>" << chat[i]->content << "<|message_sep|>";
-            }
-        }
-
-        // Add generation prompt if needed
-        if (add_ass) {
-            ss << "assistant<|role_sep|>";
-        }
-    } else {
-        // template not supported
-        return -1;
-    }
-    dest = ss.str();
-    return dest.size();
-}
-
 int32_t llama_chat_apply_template(
                 const struct llama_model * model,
                               const char * tmpl,
@@ -22651,15 +12371,15 @@ int32_t llama_chat_apply_template(
     std::string curr_tmpl(tmpl == nullptr ? "" : tmpl);
     if (tmpl == nullptr) {
         GGML_ASSERT(model != nullptr);
-        // load template from model
-        std::vector model_template(2048, 0); // longest known template is about 1200 bytes
-        std::string template_key = "tokenizer.chat_template";
-        int32_t res = llama_model_meta_val_str(model, template_key.c_str(), model_template.data(), model_template.size());
-        if (res < 0) {
+
+        // load template from model, if available
+        const auto & it = model->gguf_kv.find("tokenizer.chat_template");
+        if (it != model->gguf_kv.end() && it->second.size() > 0) {
+            curr_tmpl = it->second;
+        }
+        else {
             // worst case: there is no information about template, we will use chatml by default
-            curr_tmpl = "chatml"; // see llama_chat_apply_template_internal
-        } else {
-            curr_tmpl = std::string(model_template.data(), model_template.size());
+            curr_tmpl = "chatml";  // see llm_chat_apply_template
         }
     }
 
@@ -22671,11 +12391,11 @@ int32_t llama_chat_apply_template(
     }
 
     std::string formatted_chat;
-    llm_chat_template detected_tmpl = llama_chat_detect_template(curr_tmpl);
+    llm_chat_template detected_tmpl = llm_chat_detect_template(curr_tmpl);
     if (detected_tmpl == LLM_CHAT_TEMPLATE_UNKNOWN) {
         return -1;
     }
-    int32_t res = llama_chat_apply_template_internal(detected_tmpl, chat_vec, formatted_chat, add_ass);
+    int32_t res = llm_chat_apply_template(detected_tmpl, chat_vec, formatted_chat, add_ass);
     if (res < 0) {
         return res;
     }
@@ -22685,15 +12405,6 @@ int32_t llama_chat_apply_template(
     return res;
 }
 
-int32_t llama_chat_builtin_templates(const char ** output, size_t len) {
-    auto it = LLM_CHAT_TEMPLATES.begin();
-    for (size_t i = 0; i < std::min(len, LLM_CHAT_TEMPLATES.size()); i++) {
-        output[i] = it->first.c_str();
-        std::advance(it, 1);
-    }
-    return (int32_t) LLM_CHAT_TEMPLATES.size();
-}
-
 //
 // sampling
 //
@@ -22741,6 +12452,8 @@ int llama_split_prefix(char * dest, size_t maxlen, const char * split_path, int
 
 const char * llama_print_system_info(void) {
     static std::string s;
+    s.clear(); // Clear the string, since it's static, otherwise it will accumulate data from previous calls.
+
 
     for (size_t i = 0; i < ggml_backend_reg_count(); i++) {
         auto * reg = ggml_backend_reg_get(i);
@@ -22761,6 +12474,10 @@ const char * llama_print_system_info(void) {
     return s.c_str();
 }
 
+//
+// perf
+//
+
 struct llama_perf_context_data llama_perf_context(const struct llama_context * ctx) {
     struct llama_perf_context_data data = {};
 
@@ -22796,47 +12513,3 @@ void llama_perf_context_reset(struct llama_context * ctx) {
     ctx->t_eval_us   = ctx->n_eval = 0;
     ctx->t_p_eval_us = ctx->n_p_eval = 0;
 }
-
-// For internal test use
-const std::vector> & llama_internal_get_tensor_map(
-    struct llama_context * ctx
-) {
-    return ctx->model.tensors_by_name;
-}
-
-void llama_log_set(ggml_log_callback log_callback, void * user_data) {
-    ggml_log_set(log_callback, user_data);
-    g_logger_state.log_callback = log_callback ? log_callback : llama_log_callback_default;
-    g_logger_state.log_callback_user_data = user_data;
-}
-
-static void llama_log_internal_v(ggml_log_level level, const char * format, va_list args) {
-    va_list args_copy;
-    va_copy(args_copy, args);
-    char buffer[128];
-    int len = vsnprintf(buffer, 128, format, args);
-    if (len < 128) {
-        g_logger_state.log_callback(level, buffer, g_logger_state.log_callback_user_data);
-    } else {
-        char * buffer2 = new char[len + 1];
-        vsnprintf(buffer2, len + 1, format, args_copy);
-        buffer2[len] = 0;
-        g_logger_state.log_callback(level, buffer2, g_logger_state.log_callback_user_data);
-        delete[] buffer2;
-    }
-    va_end(args_copy);
-}
-
-void llama_log_internal(ggml_log_level level, const char * format, ...) {
-    va_list args;
-    va_start(args, format);
-    llama_log_internal_v(level, format, args);
-    va_end(args);
-}
-
-void llama_log_callback_default(ggml_log_level level, const char * text, void * user_data) {
-    (void) level;
-    (void) user_data;
-    fputs(text, stderr);
-    fflush(stderr);
-}