ngxson/llama.cpp

Pull Requests Commits

mtmd: clarify that we no longer accept AI-generated PRs

ngxson committed 42 days ago

4d8db3a1

llama_fit_params: return enum for fail vs. error (#18374)

JohannesGaessler committed 42 days ago

Verified a52dc60b

llama-fit-params: fix Gemma 3 calculation (#18372)

JohannesGaessler committed 42 days ago

Verified 9045c9af

vulkan: preprocess mul_mat_id experts and discard workgroups more quickly (#18352)

jeffbolznv committed 43 days ago

Verified c9ced491

vulkan: optimize decodeFuncB in coopmat2 mul_mat_id shader (#18349)

jeffbolznv committed 43 days ago

Verified 7ac89021

vulkan: Use BK=32 for coopmat2 mul_mat_id (#18332)

jeffbolznv committed 43 days ago

Verified 9bf20d8a

vulkan: small dequantization improvements (#18380)

netrunnereve committed 43 days ago

Verified cb999704

vulkan: Support UPSCALE w/antialias (#18327)

jeffbolznv committed 43 days ago

Verified b96b82fc

vulkan: handle rope with large number of rows (#18306)

jeffbolznv committed 43 days ago

Verified 10dc500b

server : fix crash when seq_rm fails for hybrid/recurrent models (#18391)

o7si committed 43 days ago

Verified 4893cc07

docs: added note for pre SYCL Intel hardware (#18016)

alosslessdev committed 43 days ago

Verified af3be131

CANN: implement the SSM_CONV operator (#17737)

0Marble committed 43 days ago

Verified b07cda68

ggml-cuda: fix regex for arch list (#18371)

am17an committed 44 days ago

Verified 85c40c9b

cuda: optimize cumsum cub path (#18362)

am17an committed 44 days ago

Verified 83b3b1c2

ggml-cuda: fix blackwell native builds (#18361)

am17an committed 44 days ago

Verified b0fb0f0a

CANN: Add support for CONV_TRANSPOSE_1D when kernel size > 255 (#17934)

Intellouis committed 44 days ago

Verified e68c19b0

ggml : optimize cuda cumsum fallback kernel (#18343)

Aadeshveer committed 44 days ago

Verified c54bba86

server: (router) add stop-timeout option (#18350)

ngxson committed 44 days ago

Verified f5acfb2f

model: support MiMo-V2-Flash (#18328)

ngxson committed 45 days ago

Verified 4cbafad4

fit-params : fix race condition in fit-params output (#18276)

Aadeshveer committed 45 days ago

Verified c1842842

CUDA: experimental native mxfp4 support for blackwell (#17906)

am17an committed 45 days ago

Verified c8a2417d

model : support for LlamaBidirectionalModel architecture (#18220)

sfallah committed 45 days ago

Verified 54132f1b

vulkan: fix command buffer corruption in ggml_backend_vk_event_wait (#18302)

jeffbolznv committed 45 days ago

Verified 2a9ea202

CANN : refactor ACL graph cache (#17752)

wangweixuan committed 45 days ago

Verified ce7a6dc0

docs: Fix typos in SYCL documentation (#18269)

yoka committed 45 days ago

Verified 1ce0126b

vulkan: use fewer FA rows for small cache runs (#18280)

0cc4m committed 45 days ago

Verified 7f459c98

CANN: Uses yarn_ramp cache in ROPE (#17725)

TianHao324 committed 45 days ago

Verified cf2ffc02

common: add `LLAMA_ARG_OVERRIDE_TENSOR` env var for `-ot` arg (#18267)

ddh0 committed 45 days ago

Verified 10355dc7

server: return_progress to also report 0% processing state (#18305)

ngxson committed 46 days ago

Verified 5ee4e43f

webui: apply webui_settings on first load (#18223)

ServeurpersoCom committed 46 days ago

Verified 5b6c9bc0

Older