sync : ggml#2573

Merged

ggerganov merged 41 commits intomasterfrom

Nov 20, 2024

Commits on Nov 19, 2024

scripts : update sync
ggerganov
committed
ggml : build backends as libraries (llama/10256)

committed
backend cpu: add online flow for aarch64 Q4_0 GEMV/GEMM kernels (llama/9921)

authored and
ggerganov
committed
sycl: Use syclcompat::dp4a (llama/10267)

Rbiessy
authored and
ggerganov
committed
AVX BF16 and single scale quant optimizations (llama/10212)

netrunnereve
authored and
ggerganov
committed
cmake : restore CMakeLists.txt (llama/10256)
ggerganov
committed
sync : leftovers (ggml/0)
ggerganov
committed
ggml : fix some build issues

slaren
authored and
ggerganov
committed
ggml : remove duplicated sources from the last sync (ggml/1017)
ggerganov
committed
ggml: new optimization interface (ggml/988)

JohannesGaessler
authored and
ggerganov
committed
Make updates to fix issues with clang-cl builds while using AVX512 flags (llama/10314)

Srihari-mcw
authored and
ggerganov
committed
ggml : optimize Q4_0 into Q4_0_X_Y repack (llama/10324)

eddnjjn
authored and
ggerganov
committed
vulkan: Optimize some mat-vec mul quant shaders (llama/10296)

jeffbolznv
authored and
ggerganov
committed
llamafile : fix include path (llama/0)
ggerganov
committed
ggml : fix compile warnings (llama/0)
ggerganov
committed
ggml : adapt AMX to tensor->grad removal (llama/0)
ggerganov
committed
ggml : inttypes.h -> cinttypes (llama/0)
ggerganov
committed
ggml : fix possible buffer use after free in sched reserve (llama/9930)

slaren
authored and
ggerganov
committed
CMake: default to -arch=native for CUDA build (llama/10320)

JohannesGaessler
authored and
ggerganov
committed
CUDA: remove DMMV, consolidate F16 mult mat vec (llama/10318)

JohannesGaessler
authored and
ggerganov
committed
ggml : fix undefined reference to 'getcpu' (llama/10354)

FirstTimeEZ
authored and
ggerganov
committed
metal : refactor kernel args into structs (llama/10238)
ggerganov
committed
llama : only use default buffer types for the KV cache (llama/10358)

slaren
authored and
ggerganov
committed
CMake: fix typo in comment [no ci] (llama/10360)

JohannesGaessler
authored and
ggerganov
committed
CUDA: fix MMV kernel being used for FP16 src1 (llama/10357)

JohannesGaessler
authored and
ggerganov
committed
metal : add `GGML_UNARY_OP_ELU` kernel (ggml/1018)

PABannier
authored and
ggerganov
committed
Vulkan: Fix device info output format specifiers (llama/10366)

0cc4m
authored and
ggerganov
committed
metal : fox offset integer overflows in im2col (ggml/1015)

pminev
authored and
ggerganov
committed
vulkan: remove use of null initializer (llama/10372)

jeffbolznv
authored and
ggerganov
committed
cuda : only use native when supported by cmake (llama/10389)

slaren
authored and
ggerganov
committed
sycl: Revert MUL_MAT_OP support changes (llama/10385)

Alcpz
authored and
ggerganov
committed
vulkan: Optimize soft_max (llama/10301)

jeffbolznv
authored and
ggerganov
committed
sycl : Add option to set the SYCL architecture for all targets (llama/10266)

Rbiessy
authored and
ggerganov
committed
cuda : fix CUDA_FLAGS not being applied (llama/10403)

slaren
authored and
ggerganov
committed
Add required ggml-base and backend libs to cmake pkg (llama/10407)

bandoti
authored and
ggerganov
committed
ggml : sync resolve (skip) (#0)
ggerganov
committed
sync : ggml
ggerganov
committed
talk-llama : sync llama.cpp
ggerganov
committed

Commits on Nov 20, 2024