ggerganov/llama.cpp

Pull Requests Commits

logs : reduce v2 (#25078)

ggerganov committed 2 hours ago

Verified 27c8bb4f

opencl: flash attention improvement (#25069)

wanghqc committed 9 hours ago

Verified ebd048fc

[CUDA] Added a cudaMemcpy2DAsync fast path to ggml_cuda_cpy (#25057)

gaugarg-nv committed 19 hours ago

Verified 0ed235ea

sycl : fix failed ut cases of norm (#25044)

arthw committed 22 hours ago

Verified 9bebfcb4

vulkan: fix step operator for 0 input (#25036)

0cc4m committed 23 hours ago

Verified 0b6529d8

binaries : Improve rpc-server and export-graph-ops names. (#25045)

ckastner committed 1 day ago

Verified c299a92c

ci : add windows-openvino to check-release (#25022)

CISC committed 1 day ago

Verified 0275c0f8

tests : fix test-chat-template --no-common option (#25075)

CISC committed 1 day ago

Verified 83d385b4

app : allow --version, --licenses & --help (#25054)

angt committed 1 day ago

Verified 050ee92d

sched : reintroduce less synchronizations during split compute (#20793)

aendk committed 1 day ago

Verified 3fc4e105

devops : add llama in all docker images (#25035)

angt committed 1 day ago

Verified 5d8ccdf9

arg: fix handling --spec-draft-hf and --hf-repo-v (#25043)

ngxson committed 1 day ago

Verified 024930c6

openvino: Update to OV 2026.2.1, self-contained release packages, operator improvements (#24974)

ravi9 committed 1 day ago

Verified 5397c361

ggerganov committed 1 day ago

e7ea94af

ggml : bump version to 0.15.3 (ggml/1550)

ggerganov committed 1 day ago

96183e98

vulkan: opt mul_mat_vecq for mi50 (#22933)

chraac committed 1 day ago

Verified 487a6cc1

vulkan: add INTEL_XE1 arch enum and enable coopmat1 on Intel Xe-LPG Plus (#24404)

fish-jiang committed 1 day ago

Verified 5a6a0dd7

ui: fix accessibility for hover-gated interactive elements assisted by claude(in debugging and tests) (#24727)

sanjayahari committed 1 day ago

Verified ded1561b

vulkan: Workaround compiler bug in conv2d coopmat2 path (#24924)

jeffbolznv committed 1 day ago

Verified 9df06805

CUDA: add cublasSgemmBatched mapping for HIP/MUSA vendor headers (#25033)

leonardHONG committed 1 day ago

Verified 2f18fe13

ggml-cpu: fix SVE leftover path in ggml_vec_dot_f32 (#24699)

tdakhran committed 2 days ago

Verified c16c35b8

server + ui: SSE Replay Buffer (#23226)

ServeurpersoCom committed 2 days ago

Verified 1a87dcdc

sycl : clamp softmax input to avoid underflow (#24941)

Jassieluo committed 2 days ago

Verified e7e3f350

mtmd: add more validations (#25013)

ngxson committed 2 days ago

Verified b11f7c16

CUDA: batch out_prod broadcast (dps2>1) path with cublasSgemmBatched (#24426)

leonardHONG committed 2 days ago

Verified f818065d

mamba2: remove hardcoded 2x expansion factor and invalid d_inner % d_state check (#23082)

limloop committed 2 days ago

Verified 960d628f

opencl: flush profiling batch at shutdown for incomplete batches (#25016)

shaofeiqi committed 2 days ago

Verified 5c7c22c3

xcframework : disable mtmd video on i/tv/visionos (#25018)

CISC committed 2 days ago

Verified beac5309

model : Add label for LFM2.5-230M (#25008)

tdakhran committed 2 days ago

Verified 9d5d882d

CUDA: Various fixes to `cpy.cu` (#25000)

ORippler committed 2 days ago

Verified 1ec44d17

Older