sync : ggml#3342

Merged

ggerganov merged 38 commits intomasterfrom

sync-ggml-25-07-28

Jul 28, 2025

Commits on Jul 28, 2025

ggml-cpu : remove stdlib include from repack.cpp (ggml/1276)

danbev
authored and
ggerganov
committed
cmake : fix usage issues (ggml/1257)

dg0yt
authored and
ggerganov
committed
Vulkan: Fix fprintf format-security warning (llama/14770)

0cc4m
authored and
ggerganov
committed
vulkan: Add logging for bf16 features to ggml_vk_print_gpu_info (#13274) (llama/14707)

Peter0x44
authored and
ggerganov
committed
ggml: adds CONV_2D op and direct GEMM Vulkan implementation (llama/14316)

etasnadi
authored and
ggerganov
committed
vulkan/cuda: Fix im2col when KW!=KH (llama/14789)

jeffbolznv
authored and
ggerganov
committed
kleidiai: add support for get_rows (llama/14676)

chaxu01
authored and
ggerganov
committed
sycl: Fix im2col (llama/14797)

Rbiessy
authored and
ggerganov
committed
opencl: add conv2d kernel (llama/14403)

rmatif
authored and
ggerganov
committed
opencl: fix `im2col` when `KW!=KH` (llama/14803)

CISC
authored and
ggerganov
committed
cuda: remove linking to cublasLt (llama/14790)

yeahdongcn
authored and
ggerganov
committed
opencl: remove unreachable `return` (llama/14806)

lhez
authored and
ggerganov
committed
cuda : implement bf16 cpy ops and enable bf16 cont (llama/14763)

CISC
authored and
ggerganov
committed
vulkan: fix rms_norm_mul to handle broadcasting dim0 (llama/14817)

jeffbolznv
authored and
ggerganov
committed
CUDA: add fused rms norm (llama/14800)

am17an
authored and
ggerganov
committed
CANN: weight format to NZ for Ascend310P3 (llama/14407)

tqgy6
authored and
ggerganov
committed
ggml: fix loongarch quantize_row_q8_1 error (llama/14827)

lixing-star
authored and
ggerganov
committed
CUDA: fix quantized KV cache + multiple sequences (llama/14822)

JohannesGaessler
and
ggerganov
committed
CUDA: fix compilation with GGML_CUDA_F16 (llama/14837)

JohannesGaessler
authored and
ggerganov
committed
CUDA: fix overflow in FA, tune performance (llama/14840)

JohannesGaessler
authored and
ggerganov
committed
sycl: fix undefined variable in work group size check (llama/14843)

djeong20
authored and
ggerganov
committed
metal : fix fusion across different encoders (llama/14849)
ggerganov
committed
sycl: fixed semantics of block offset calculation (llama/14814)

Alcpz
authored and
ggerganov
committed
cmake : Indent ggml-config.cmake (ggml/1310)

dg0yt
authored and
ggerganov
committed
musa: upgrade musa sdk to rc4.2.0 (llama/14498)

yeahdongcn
authored and
ggerganov
committed
sched : fix multiple evaluations of the same graph with pipeline parallelism (llama/14855)

slaren
authored and
ggerganov
committed
rpc : check for null buffers in get/set/copy tensor endpoints (llama/14868)

struct
authored and
ggerganov
committed
ggml : remove invalid portPos specifiers from dot files (llama/14838)

ORippler
authored and
ggerganov
committed
opencl: add fused `rms_norm_mul` (llama/14841)

lhez
authored and
ggerganov
committed
metal: SSM_SCAN performance (llama/14743)

gabe-l-hart
and
ggerganov
committed
ggml-cpu : disable GGML_NNPA by default due to instability (llama/14880)

taronaeo
authored and
ggerganov
committed
musa: fix build warnings (unused variable) (llama/14869)

yeahdongcn
authored and
ggerganov
committed
CANN: Implement GLU ops (llama/14884)

hipudding
authored and
ggerganov
committed
HIP: Enable Matrix cores for MMQ Kernels, Enable stream-K for CDNA 3 (llama/14624)

deepsek
authored and
ggerganov
committed
vulkan: skip empty set_rows to avoid invalid API usage (llama/14860)

jeffbolznv
authored and
ggerganov
committed
vulkan : add fp16 support for the conv_2d kernel (llama/14872)

Green-Sky
authored and
ggerganov
committed
sync : ggml
ggerganov
committed
talk-llama : sync llama.cpp
ggerganov
committed