ggml-org/llama.cpp

Pull Requests Commits

models : consolidate nextn row cropping and make graph topology independent of embeddings_nextn

ggerganov committed 2 hours ago

57f21560

mimo2 : always emit h_nextn to keep graph topology independent of nextn extraction

ggerganov committed 3 hours ago

aaf139bc

hexagon: matmul and flash-atten scalability updates (#29974)

max-krasnyansky committed 5 hours ago

Verified 8345f333

llama.cpp : bump version to 0.6.0 (#29997)

ggerganov committed 6 hours ago

Verified d8123504

common, server : report model input/output modalities in GET /models (#29987)

angt committed 6 hours ago

Verified 4d60b4d0

ggerganov committed 7 hours ago

c06f8416

ggml : bump version to 0.26.0 (ggml/1652)

ggerganov committed 7 hours ago

f05c8b27

CUDA: make the alloc_deps check batch independent (#29986)

am17an committed 8 hours ago

Verified e117148a

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

0cc4m committed 8 hours ago

Verified 6c59c400

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

virajwad committed 8 hours ago

Verified 3c9e747f

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

ServeurpersoCom committed 8 hours ago

Verified b809b886

ci : add "Require Docker" flag to make-release workflow (#29989)

ggerganov committed 8 hours ago

Verified 994e8f22

webui: Use toLocaleString() format consistently across chat message statistics (#27990)

grivera64 committed 8 hours ago

Verified 9d853bb3

ci : disable failing test on virtual Metal device (#29993)

ggerganov committed 9 hours ago

Verified 8f9ae20c

server: support vision input for Clef (#29969)

ngxson committed 9 hours ago

Verified 9871df59

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

kmorennv committed 10 hours ago

Verified 8b2fbaf3

ci : disable unused qemu in docker build (#29984)

CISC committed 11 hours ago

Verified 2ed93db4

server: reject partial media truncation (#24076)

he-yufeng committed 12 hours ago

Verified 8e164219

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

fxgsell committed 13 hours ago

Verified 806eee98

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

fxgsell committed 13 hours ago

Verified b3daa077

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

ggerganov committed 13 hours ago

Verified c173a53b

kv-cache: fix restoring mismatched KV cache rotation by saving exact rotation metadata (#28498)

eapache committed 13 hours ago

Verified 21079106

ci : winget urls must be separate strings (#29978)

CISC committed 14 hours ago

Verified e5983d67

ci : fix docker workflow permissions (#29979)

CISC committed 14 hours ago

Verified 4ca6b76f

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

alanhc committed 14 hours ago

Verified 9f12cd4a

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

EAddario committed 14 hours ago

Verified ebe18bee

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

yomaytk committed 14 hours ago

Verified 8216c846

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

ServeurpersoCom committed 15 hours ago

Verified 1b43d311

metal : few-row MMA mat-mul (#29869)

pratiknarola-t committed 16 hours ago

Verified a3a1c474

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

ynankani committed 16 hours ago

Verified 9d3aba6b

Older