ggml-org/llama.cpp

Pull Requests Commits

convert: add support for mtp flags

am17an committed 30 days ago

a0484817

convert: fix mtp conversion

am17an committed 30 days ago

33a64b17

fix v granularity

am17an committed 30 days ago

c645b169

llama : fix tensor split for fused qkv with uneven K/V head sizes

am17an committed 30 days ago

e52cdd8c

cuda : add conv3d with implicit GEMM (#29137)

leejet committed 30 days ago

Verified 53ed051c

model : add Ling 3.0 VL support (#29151)

aetherbird committed 30 days ago

Verified f830688e

server,common : fix the GCC 12 stringop-overread false positive (again) (#29325)

angt committed 30 days ago

Verified 2b705839

test-save-load-state : print a per-model results table in --models mode (#29316)

ggerganov committed 30 days ago

Verified 4c5957c2

hexagon: reject MUL_MAT_ID when src1 precision is F32 (#29348)

jhen0409 committed 30 days ago

Verified 9710a321

scripts : make-release-desc - link previous release in changelog title (#29336)

ggerganov committed 30 days ago

Verified 013b31c0

convert : allow vision target for DFlash/Dspark (#29339)

tdakhran committed 30 days ago

Verified bd4f514d

server: allow preset to set log file (#29334)

ngxson committed 30 days ago

Verified b9ae43a5

tests: add `-b/--backend` option to test-llama-archs for testing a specific backend (#27372)

yomaytk committed 30 days ago

Verified d2e54583

ci : use hf-jobs-cpu-xl runner in server sanitize workflow (#29297)

ggerganov committed 30 days ago

Verified 6e60f356

opencl: add A8 Q6_K non-MoE dp4a binary kernel (#29057)

shaofeiqi committed 30 days ago

Verified fee39dd9

llama.cpp : bump version to 0.5.0 (#29333)

ggerganov committed 30 days ago

Verified 7fe450e1

ggerganov committed 30 days ago

177cd8cc

ggml : bump version to 0.25.1 (ggml/1637)

ggerganov committed 30 days ago

e4e2f623

CUDA: add a reserve to avoid spurious warning on older GCC builds (#29317)

am17an committed 30 days ago

Verified 66fba63a

common : keep HF cache dir as path, expose UTF-8 only for logs (#29320)

angt committed 30 days ago

Verified bddf8263

metal: add the missing f32 x bf16 mul_mv variants (#28741)

ServeurpersoCom committed 30 days ago

Verified 95753896

CUDA: enable sparse-fa for dsv4 prefill (again) (#29298)

am17an committed 30 days ago

Verified dc9879cf

server: fix token counting API crash on sleep (#29309)

willweimike committed 31 days ago

Verified 42916d83

jinja : parse unary +/- before variables (#29244)

cs-fisha committed 31 days ago

Verified 4e416ee7

metal : key the fa-vec tuned table by family instead of SKU (#29075)

forforever73 committed 31 days ago

Verified ee3ecce0

server: accept OpenAI video_url content type and data: video URIs (#27921)

calebrio02 committed 31 days ago

Verified 057494f9

server: Dedup the draft HF model via dedup-cache-models (#27934)

DreamingWater committed 31 days ago

Verified bcbc936a

ci : fix build-cmake runner target (#29299)

CISC committed 31 days ago

Verified 26758d38

model-conversion : add causal-compare-logits recipe (#29305)

danbev committed 31 days ago

Verified 18f9f7be

model : support Gemma4 DSpark draft backbone (#29226)

hthadicherla committed 31 days ago

Verified 633733d0

Older