ggml-org/llama.cpp

Pull Requests Commits

llama: add lazy mode auto, fix iGPU regression

0cc4m committed 11 hours ago

38d20da3

mtmd: propagate const to preproc class (#28310)

ngxson committed 14 hours ago

Verified de8656bd

metal : add sparse FA (#28098)

ggerganov committed 14 hours ago

Verified 7bb0fc18

metal : fix glu dispatch with ne00 = 1 (#28306)

ggerganov committed 14 hours ago

Verified 0df017d6

mtmd : add `const` in various places (#28307)

madsmtm committed 15 hours ago

Verified f45576aa

CUDA: Allow concurrent streams per split for multi-GPU (#28198)

tannerbruhn committed 15 hours ago

Verified 0ba6499c

vulkan: fix FA dequant path engagement (#28190)

Nathanw1014 committed 16 hours ago

Verified c7bda030

sycl : enhance the api to support peer-to-peer copy (#27550)

arthw committed 17 hours ago

Verified 0df974d7

convert : skip bias_vl tensor in DeepSeek-V4 DSpark conversion (#28294)

ggerganov committed 17 hours ago

Verified d646c9d1

misc : prevent RAM peaking at model loading stage (#27483)

tdakhran committed 17 hours ago

Verified 5ec4eab6

sycl: reduce redundant work in Q4_K multi-column MMVQ (#27062)

Eurekatic committed 18 hours ago

Verified 4aa6ffba

model: add NVIDIA Nemotron-3-Puzzle-75B-A9B (NemotronHPuzzle) support (#25444)

YanissAmz committed 18 hours ago

Verified c61b98b8

mtmd: fix idefics3 preproc (#28273)

ngxson committed 1 day ago

Verified 67a17c17

finetune: fix no KV cache (#27199)

ngxson committed 1 day ago

Verified 159b7414

server : accept data: URLs for input_video and input_audio (#27735)

geckguy committed 1 day ago

Verified 9cffdcc8

ggml-hexagon: add F16 support for unary ops (#28228)

cqderek committed 1 day ago

Verified f027c4f1

mtmd: add mtmd_tokenize_from_parts() (#28250)

ngxson committed 1 day ago

Verified 73390547

metal : add fa-vec tunings for M3 (#28236)

init-22 committed 1 day ago

Verified 9cc33944

metal : fix memory query under low-memory conditions (#27701)

madsmtm committed 1 day ago

Verified 8c0b9cd0

ci : check for missing autoreleasepools (#27884)

nikwen committed 1 day ago

Verified 03dbcc53

Update ROCm to 10.0.0 release (#27803)

superm1 committed 1 day ago

Verified cff18443

model: correctly support input vision for deepseek4 (#28154)

ngxson committed 1 day ago

Verified 9400c894

ci : enable hf-jobs on server-cuda (#28258)

CISC committed 1 day ago

Verified d5fec32a

ggml-cuda : remove unused vars (#28235)

angt committed 1 day ago

Verified 3d3d7c81

common, server : enable preserve_reasoning kwarg by default, log its effective state (#28174)

ggerganov committed 1 day ago

Verified e750b887

mtmd: support DeepSeek-V4-Flash-Vision-Exp (#28133)

ngxson committed 1 day ago

Verified 7798007a

CUDA + ggml: add sparse-fa for DSV4/GLM (#27970)

am17an committed 1 day ago

Verified 8e93a977

mtmd: Fix Qwen3-tts-0.6b (#28231)

ServeurpersoCom committed 1 day ago

Verified 0f3a71be

ggml: avoid KleidiAI buffer type init on dispatch (#27891)

ac-mmi committed 1 day ago

Verified b81c99b4

hexagon: MUL_MAT and MUL_MAT_ID fusion and fixes (#28202)

max-krasnyansky committed 1 day ago

Verified 960dffab

Older