ggerganov/llama.cpp

Pull Requests Commits

Copilot committed 47 days ago

Verified 7cce8904

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

ggerganov committed 47 days ago

Verified e750b887

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

ngxson committed 47 days ago

Verified 7798007a

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

am17an committed 47 days ago

Verified 8e93a977

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

ServeurpersoCom committed 47 days ago

Verified 0f3a71be

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

ac-mmi committed 47 days ago

Verified b81c99b4

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

max-krasnyansky committed 47 days ago

Verified 960dffab

vulkan: handle larger batch sizes (>4) efficiently for IQ3_S mat-vec (#27449)

LaurentZuijdwijk committed 47 days ago

Verified ba8818cb

vulkan : only request VK_KHR_shader_bfloat16 extension if supported (#28155)

madsmtm committed 47 days ago

Verified 56dd8150

ggml-cpu : conditionally add SpacemiT IME kernel sources (#27961)

alanhc committed 47 days ago

Verified 2637dfe3

opencl: fix out‐of‐bound reads in the Adreno image kernels (#27632)

wanghqc committed 48 days ago

Verified 43d87ff2

hexagon: add missing FARF logs for cpy/get_rows/set_rows/gdn ops (#28217)

trivikram-reddy1 committed 48 days ago

Verified 69320fef

metal : add metallib build support for xcframework (#28163)

jhen0409 committed 48 days ago

Verified b96806d9

cuda: fuse MoE weighted expert reduction (#25952)

anujj committed 48 days ago

Verified 3466812d

kv-cells: look up the n-gram history in the sequence position index (#28040)

ServeurpersoCom committed 48 days ago

Verified b356fa26

context : autoscale n_ctx_train when yarn scaling specified (#28030)

CISC committed 48 days ago

Verified dfc29b64

models : appropriately flag noscan ssm_a tensors (#28121)

CISC committed 48 days ago

Verified f28493c7

model : fix gemma4-assistant (#28183)

CISC committed 48 days ago

Verified 73159c30

model : load relevant arrays with n_layer_all (#28173)

CISC committed 48 days ago

Verified d11b3cc7

Revert "sycl : add Kronecker product FWHT support for sizes 384, 640, 768, 12…" (#28184)

Titaniumtown committed 48 days ago

Verified c845263f

sycl : add Kronecker product FWHT support for sizes 384, 640, 768, 1280 (#28016)

philip-jingxin committed 48 days ago

Verified 1f3d3187

metal : add fa-vec tuning for M2 Pro (#28122)

lstolcman committed 48 days ago

Verified 8887a48f

metal : add fa-vec tunings for A18 Pro (MacBook Neo) (#28152)

jhen0409 committed 48 days ago

Verified be789c34

model : load hparams.n_layer_nextn before n_layer() calls (#28159)

CISC committed 48 days ago

Verified 9d817213

metal : fix more leaks due to missing autoreleasepools (#27883)

nikwen committed 48 days ago

Verified fe2120bc

metal : add fa-vec tuning for M2 Max (#28015)

ggerganov committed 48 days ago

Verified d08c7872

sycl : support limit max alloc memory within 2GB for host-pinned memory (#27559)

arthw committed 48 days ago

Verified 5eec3ad0

qwen4exp: fix seq_cp, block position keying, mtmd input, cuda abort, add tests (#27941)

danielhanchen committed 48 days ago

Verified 36b10154

tests : fix log verbosity for test-llama-archs (#28147)

ggerganov committed 48 days ago

Verified d086dbb3

quantize: row-slab stream to avoid thread starvation (#27830)

ngxson committed 48 days ago

Verified 1b89a43e

Older