sync : ggml#3300

Merged

ggerganov merged 35 commits intomasterfrom

sync-ggml-25-07-01

Jul 1, 2025

Commits on Jul 1, 2025

ggml-quants : rename best_mad to best_error (ggml/1283)

danbev
authored and
ggerganov
committed
ggml-cpu : "align corners" for bilinear upscale/downscale (ggml/1285)

Acly
authored and
ggerganov
committed
metal : fix thread-safety (llama/14300)
ggerganov
committed
Add support for VK_EXT_debug_utils to add labels to Vulkan objects. (llama/13792)

mtavenrath
authored and
ggerganov
committed
CUDA: add mean operation (llama/14313)

am17an
authored and
ggerganov
committed
HIP: enable vec fattn on RDNA4 (llama/14323)

IMbackK
authored and
ggerganov
committed
CUDA: mul_mat_v support for batch sizes > 1 (llama/14262)

JohannesGaessler
authored and
ggerganov
committed
CUDA/HIP: optimize mmv paths taken for HIP devices (llama/14324)

authored and
ggerganov
committed
opencl: ref count `ggml_backend_opencl_context` and refactor profiling (llama/14254)

lhez
authored and
ggerganov
committed
sycl: GGML_SYCL_DISABLE_OPT on by default for all Intel Devices (llama/13973)

ShanoToni
authored and
ggerganov
committed
ggml : do not output unprintable characters on GGUF load failure (llama/14381)

CISC
authored and
ggerganov
committed
ggml-cpu: enable IBM NNPA Vector Intrinsics (llama/14317)

authored and
ggerganov
committed
musa: enable fp16 mma (all) and cublas on qy2 (llama/13842)

authored and
ggerganov
committed
metal : batch rows copy in a single threadgroup (llama/14384)
ggerganov
committed
metal : add special-case mat-vec mul for ne00 == 4 (llama/14385)
ggerganov
committed
cmake: regen vulkan shaders when shaders-gen sources change (llama/14398)

bandoti
authored and
ggerganov
committed
ggml : add ggml_set_rows (llama/14274)

rgerganov
and
ggerganov
committed
vulkan: Fix GGML_VULKAN_SHADER_DEBUG_INFO (llama/14427)

jeffbolznv
authored and
ggerganov
committed
fix async_mode bug (llama/14432)

bachelor-dou
authored and
ggerganov
committed
vulkan: lock accesses of pinned_memory vector (llama/14333)

jeffbolznv
authored and
ggerganov
committed
vulkan: handle noncontig in the final case of ggml_vk_get_cpy_pipeline (llama/14378)

jeffbolznv
authored and
ggerganov
committed
CUDA: add bf16 and f32 support to cublas_mul_mat_batched (llama/14361)

am17an
authored and
ggerganov
committed
vulkan: Add fusion support for RMS_NORM+MUL (llama/14366)

authored and
ggerganov
committed
ggml : implement REGLU/GEGLU/SWIGLU ops (llama/14158)

committed
ggml : fix unmerged GGML_FPxx_TO_FPxx refactoring (llama/14443)

CISC
authored and
ggerganov
committed
SYCL: disable faulty fp16 exp kernel (llama/14395)

qnixsynapse
authored and
ggerganov
committed
scripts : make the shell scripts cross-platform (llama/14341)

vedranmiletic
authored and
ggerganov
committed
cmake : Remove redundant include path in CMakeLists.txt (llama/14452)

xiaobing318
authored and
ggerganov
committed
ggml-cpu: sycl: Re-enable exp f16 (llama/14462)

Rbiessy
authored and
ggerganov
committed
metal : disable fast-math for some cpy kernels (llama/14460)
ggerganov
committed
Add Conv2d for CPU (llama/14388)

am17an
authored and
ggerganov
committed
opencl : add GEGLU, REGLU, SWIGLU (llama/14456)

lhez
authored and
ggerganov
committed
ggml : remove trailing whitespace (llama/0)
ggerganov
committed
talk-llama : sync llama.cpp
ggerganov
committed
sync : ggml
ggerganov
committed