Allow multiple copy function pointers for CUDA graph kernel param updates (llama/7565)
143f6df
agray3
commited on
Spaces:
Sleeping
Sleeping
Commit History
Fix q_xxs using mul_mat_q (llama/7459)
0be4f48
AidanBeltonS
commited on
Add freq factors (llama/7495)
340b830
AidanBeltonS
commited on
metal : add GGML_OP_REPEAT kernels (llama/7557)
0534b5d
metal : disable FA kernel for HS=256 (llama/7556)
0c32e28
ggml : restore ggml_rope_xpos_inplace (ggml/0)
0641dee
ggml: aarch64: SVE kernels for q8_0_q8_0, q4_0_q8_0 vector dot (llama/7433)
51f504f
Masaya, Kato
commited on
ggml : silence UB sanitizer error during iq2_xxs quantization (llama/0)
9f41704
ggml : remove ggml_flash_attn and ggml_flash_ff (llama/7463)
4005bca
ggml : drop support for QK_K=64 (llama/7473)
8737d46
Update vulkan rope implementation to support frequency factors (llama/7475)
be0ec58
CUDA: fix FA out-of-bounds reads (llama/7479)
b38d0f9
CUDA: fix FA out-of-bounds writes (llama/7465)
2e26e3a
cuda : fix compile warning (llama/7454)
58db6c8
CUDA: remove incorrect precision check (llama/7454)
eb4b5e0
cuda : fix rope + add tests (llama/7452)
215ce5c
llama : add phi3 128K model support (llama/7225)
ef68527
metal : handle F16 inf values, fix FA partial offload (llama/7434)
8d153a7
CUDA: fix unused warning in mmq.cu (llama/7442)
f16510d
CUDA: deduplicate mmq code (llama/7397)
e7b20b1
rpc : track allocated buffers (llama/7411)
925eb7a
Update SYCL upscale operation (llama/7321)
3984ba6
AidanBeltonS
commited on
ggml-opencl, llama: using reserve() if count already known (llama/7272)
8325ed5
ggml : add loongarch lsx and lasx support (llama/6454)
9794ea7
junchao-loongson
Jinyang He
commited on
Add provisions for windows support for BF16 code including CMake provision for enabling AVX512_BF16 (llama/7258)
cf52931
Srihari-mcw
commited on
Vulkan Embedding Fix (llama/7360)
2bfeba3
ggml : fix another case of quants nans (llama/7387)
645c367
slaren
commited on
ggml: implement quantized KV cache for FA (llama/7372)
aef1b4b
cuda : clear error after buffer allocation failure (llama/7376)
b7f6691
slaren
commited on
Capture CUDA logging output (llama/7298)
3519475
fraxy-v
slaren
commited on
android : use "ci-android" branch for CI (llama/7341)
ff9d573
CUDA: deduplicate FlashAttention code (llama/7352)
65ab3e8
cuda : add half2 __shfl_xor() for ROCm 5.5 (llama/7263)
ad83dfd
Engininja2
commited on
Update and fix Vulkan soft_max and argsort implementations (llama/7237)
a0218a3
ggml : fix quants nans when all the group weights are very close to zero (llama/7313)
b57bcbc
slaren
commited on
CUDA: faster large batch FA without tensor cores (llama/7314)
a6d9f2d
rpc : set SO_REUSEADDR for the server socket (llama/7320)
195fe29
ggml-quants, llama : removed excess checks (llama/7274)
142d95e
ggml : rewrite silu and softmax for cpu (llama/7154)
c78b872
Justine Tunney
commited on
rpc : add command line arg for specifying backend memory
b441739
Add support for properly optimized Windows ARM64 builds with LLVM and MSVC (llama/7191)
c917076
ggml : use dynamic thread scheduling for matrix multiplication (llama/6915)
6f8daf7
kunnis
commited on
Avoid unnecessarily disabling CUDA graphs (llama/7302)
4816f6a
agray3
commited on
ggml : tag ggml_tensor::backend as deprecated (llama/7290)
1a5606e
slaren
commited on
Add missing " (llama/7303)
2c417da
AidanBeltonS
commited on
