ggml-org/ggml

Pull Requests Commits

scripts : update release summary prompt

ggerganov committed 5 days ago

Verified ffa4e8b8

ggml : bump version to 0.26.0 (#1652)

ggerganov committed 5 days ago

Verified d7cb5741

sync : llama.cpp

ggerganov committed 5 days ago

Verified 77f2f491

CUDA: make the alloc_deps check batch independent (llama/29986)

am17an committed 5 days ago

Verified 316f7e78

vulkan: fix Flash Attention shmem write out of bounds (llama/29988)

0cc4m committed 5 days ago

Verified 9e974161

vulkan: revert mul_mat_id tile selection PR #29182 (llama/29936)

virajwad committed 5 days ago

Verified eb9fa483

cuda: use the vector lightning indexer kernel on MUSA (llama/29990)

ServeurpersoCom committed 5 days ago

Verified ab367edb

CUDA: Optimize accumulation in mmq for NVFP4 type (llama/29857)

kmorennv committed 5 days ago

Verified 4c901461

vulkan: fix stale prealloc_y reuse across flash attention and soft_max (llama/29591)

fxgsell committed 5 days ago

Verified 4bb5b76c

vulkan: sparse flash attention for quantized K/V (llama/29639)

fxgsell committed 5 days ago

Verified 4c526d30

llama : fix unexpected graph reallocation in the k-pool models (llama/29958)

ggerganov committed 5 days ago

Verified 979ff228

ggml-cpu : add Q8_0 IME1 matrix kernel for SpacemiT X60 (llama/28479)

alanhc committed 5 days ago

Verified 28fb752c

vulkan : Fix undeclared identifiers when -DGGML_VULKAN_RUN_TESTS=ON (llama/29912)

EAddario committed 5 days ago

Verified f87a8749

webgpu: add MMVQ support for Q1_0/Q5_0/Q5_1/Q3_K/Q5_K/Q6_K/MXFP4 (llama/29483)

yomaytk committed 5 days ago

Verified d5486c5c

cuda: tile the lightning indexer over keys and tokens for 4 heads (llama/29901)

ServeurpersoCom committed 5 days ago

Verified c26d651d

metal : few-row MMA mat-mul (llama/29869)

pratiknarola-t committed 5 days ago

Verified e80a369b

CUDA: use MMVF for thin f16/bf16 mul_mat at small batch size (llama/29633)

ynankani committed 5 days ago

Verified 046df6f9

CUDA: prefer whole-tile FlashAttention scheduling for efficient two-stage kernels (llama/29435)

anujj committed 5 days ago

Verified 64b64bcc

CUDA: refactor swizzling code (llama/29612)

JohannesGaessler committed 5 days ago

Verified 4ad8c550

sync : llama.cpp

ggerganov committed 5 days ago

Verified ce8ec243

ggml-cpu: support BF16/FP16/FP32 K tails in tinyBLAS on x86 (llama/29806)

SongXiaoXi committed 5 days ago

Verified ebefd1bb

cuda : move neu_padded to where it is used (llama/29940)

angt committed 5 days ago

Verified 9dd7f1fb

cuda : move blocks_per_col to where it is used (llama/29939)

angt committed 5 days ago

Verified 799079d6

CUDA: fix MMQ memory fault if n_expert >> n_ubatch (llama/29941)

JohannesGaessler committed 5 days ago

Verified 51f4e809

vulkan: fix rdna4 mat_vec tuning (llama/29934)

0cc4m committed 5 days ago

Verified 8b304126

webgpu: add f16 support to fill/set_rows (llama/29897)

yomaytk committed 5 days ago

Verified be35ddc7

ci: fix flaky ADD_ADD f16 by using the fused ADD tolerance (llama/29904)

ServeurpersoCom committed 5 days ago

Verified 51aab529

ggml-openvino: update to 2026.4.1, optimize performance, expand ops, improve device listing. (llama/29852)

ravi9 committed 5 days ago

Verified 049c181a

qwen4exp : halve the indexer score memory (llama/29825)

ServeurpersoCom committed 5 days ago

Verified 84ae6f20

CUDA: fuse shared experts into MMVQ (llama/29184)

am17an committed 5 days ago

Verified e45b9044

Older