sync : ggml#2868

Merged

ggerganov merged 38 commits intomasterfrom

sync-ggml-25-03-08

Mar 8, 2025

Commits on Mar 8, 2025

Told cmake to install ggml-cpp.h as a public header file. (ggml/1126)

authored and
ggerganov
committed
cuda/cpu: Increase support for fp16 unary operations (ggml/1125)

cmdr2
authored and
ggerganov
committed
cuda/vulkan: specify fp32-only support for some operations in supports_op (ggml/1129)

cmdr2
authored and
ggerganov
committed
cuda: unary ops as float + de-duplicate (ggml/1130)

cmdr2
authored and
ggerganov
committed
ggml-cpu: Fix build with sve (llama/12059)

MollySophia
authored and
ggerganov
committed
vulkan: fix assertion when qy_needs_dequant (llama/12068)

jeffbolznv
authored and
ggerganov
committed
cmake: Fix ggml backend dependencies and installation (llama/11818)

vvuksanovic
authored and
ggerganov
committed
vulkan: improve im2col (llama/11826)

daniandtheweb
authored and
ggerganov
committed
vulkan: matmul dequantization improvements (llama/12015)

netrunnereve
authored and
ggerganov
committed
CANN: Fix build error with GCC 13 (llama/11990)

hipudding
authored and
ggerganov
committed
ggml: aarch64: implement SVE kernels for q2_k_q8_k vector dot (llama/12064)

authored and
ggerganov
committed
CUDA: fix logic for V100 + GGML_CUDA_FORCE_MMQ (llama/12098)

JohannesGaessler
authored and
ggerganov
committed
vulkan: add specific MMV kernels for IQ2 and IQ3 quants + optimizations (llama/11595)

remyoudompheng
authored and
ggerganov
committed
ggml : upgrade init_tensor API to return a ggml_status (llama/11854)

authored and
ggerganov
committed
CUDA: compress mode option and default to size (llama/12029)

Green-Sky
authored and
ggerganov
committed
ggml-backend : keep paths in native string type when possible (llama/12144)

slaren
authored and
ggerganov
committed
SYCL: Move CPY kernels to a separate file and add few missing kernels (llama/12133)

qnixsynapse
authored and
ggerganov
committed
ggml : fix kleidiai build (llama/12159)

ag2s20150909
authored and
ggerganov
committed
HIP: implement FlashAttention via rocWMMA for CDNA and RDNA3+ (llama/12032)

hjc4869
authored and
ggerganov
committed
ggml : portability fixes for VS 2017 (llama/12150)

authored and
ggerganov
committed
vulkan : sync (llama/0)
ggerganov
committed
ggml : ggml_compute_forward_concat() for arbitrary tensor type (ggml/1118)

vmobilis
authored and
ggerganov
committed
ggml : fix GGMLMetalClass ODR (llama/12200)

pminev
authored and
ggerganov
committed
SYCL: Disable f16 Unary OPs as not supported by the kernels (llama/12201)

qnixsynapse
authored and
ggerganov
committed
ggml-cpu: Faster IQ1 mul_mat_vec on AVX2 using BMI2 instructions (llama/12154)

remyoudompheng
authored and
ggerganov
committed
opencl : fix profile-related errors (llama/12095)

authored and
ggerganov
committed
opencl : fix `ulong` kernel args were set from `int` variables (llama/12174)

linehill
authored and
ggerganov
committed
opencl : fix buffer alignment (llama/12197)

linehill
authored and
ggerganov
committed
HIP/CUDA: set the paramerter value in maintain_cuda_graph instead of replaceing it. (llama/12209)

IMbackK
authored and
ggerganov
committed
CUDA: fix FA logic for PTX 7.0 and CC >= 7.5 (llama/12222)

JohannesGaessler
authored and
ggerganov
committed
cmake : fix undefined reference errors for std::filesystem in ggml (#12092) (llama/12094)

authored and
ggerganov
committed
opencl: Noncontiguous `norm`, `rms_norm`, disable `fp16` for some ops (llama/12217)

lhez
authored and
ggerganov
committed
metal : fix default.metallib build (llama/12224)

danbev
authored and
ggerganov
committed
metal : simplify kernel arguments using a struct (ggml/3229) (llama/12194)

authored and
ggerganov
committed
ggml-cpu: faster AVX2 variant for IQ1_M (llama/12216)

remyoudompheng
authored and
ggerganov
committed
sync : ggml
ggerganov
committed
cmake : fix ggml-config (ggml/0)
ggerganov
committed
objc : fix build, tmp remove GPU support, use C++17
ggerganov
committed