whisper : update CMakeLists.txt to handle deprecated gpu Warnings (#3163)

ruby : add GGML_SYCL_DNN option to ruby bindings (#3172)

talk-llama : sync llama.cpp

sync : ggml

CANN: Support MOE Model MUL_MAT_ID (llama/13042)

cmake: use the current build config for vulkan-shaders-gen (llama/13595)

vulkan: move common FA code to flash_attn_base.comp (llama/13556)

vulkan: use scalar FA rather than coopmat2 when N==1 (llama/13554)

metal : add FA-vec kernel for head size 64 (llama/13583)

sycl : fixed compilation warnings (llama/13582)

gguf : use ggml log system (llama/13571)

sycl: simplify bin_bcast_kernel (llama/13383)

sycl: reordered Q4_K MMVQ (llama/13109)

sycl: use oneDNN for matrices multiplication (llama/12972)

arm64: optimize q6_k_q8_k kernel with i8mm (llama/13519)

CUDA: fix crash on large batch size for quant. MoE (llama/13537)

CUDA: faster Deepseek FA, add Turing support (llama/13435)

cmake: simplify vulkan shader test logic (llama/13263)

vulkan: KHR_coopmat flash attention (llama/13506)

vulkan: workaround FA compile failures on macos (llama/13517)

metal : use FA-vec kernel up to batch size 20 (llama/13496)

metal : optimize multi-sequence FA vec kernel (llama/13493)

ggml-cpu: Update KleidiAI to v1.6 and fix include directives (llama/13509)

mnist: fix segmentation fault (ggml/1227)

ggml : fix apple OS check in ggml_print_backtrace (ggml/1229)

ggml : Fix missing backtrace on Linux (ggml/1228)

examples : add vad-speech-segments to win warns [no ci] (#3170)

vad : return early if no vad segments are detected (#3158)

vad : store VAD context in whisper_state (#3156)

whisper : add build_*/ to .gitignore [no ci] (#3157)

examples : add --print-confidence option to cli (#3150)

vad : add download-vad-model scripts (#3149)

server : add --flash-attn usage output (#3152)

talk-llama : sync llama.cpp

whisper : update to ggml-backend changes (#0)

sync : ggml

ggml : add mrope kernel for metal (llama/13457)

metal : optimize MoE for large batches (llama/13388)

opencl: remove unnecessary assert for `add` (llama/13257)

llama/ggml: add LLM training support (llama/10544)

ggml-cpu: Integrate fp32=bf16xbf16 SME KleidiAI kernel (llama/13053)

CUDA: fix misaligned synchronization in FA (llama/13469)

enable dpcpp nightly builds with libraries (llama/13406)

CUDA: fix crash with partial offloading of MoE (llama/13439)

Add `--no-op-offload` to improve `-ot` pp perf in MoE models like llama4 400B (llama/13386)

CUDA: fix race conditions FlashAttention kernels (llama/13438)

CUDA: fix FlashAttention on Turing (llama/13415)

vulkan: scalar flash attention implementation (llama/13324)

sycl : implementation of reordered Q4_0 MMVQ for Intel GPUs (llama/12858)

CUDA: FA support for Deepseek (Ampere or newer) (llama/13306)