ggml-org/llama.cpp

Pull Requests Commits
ggml-cuda: Use same regex for GGML_NATIVE=OFF (#18407)
am17an committed 31 days ago
Verified 06705fdc
llama_fit_params: return enum for fail vs. error (#18374)
JohannesGaessler committed 31 days ago
Verified a52dc60b
llama-fit-params: fix Gemma 3 calculation (#18372)
JohannesGaessler committed 31 days ago
Verified 9045c9af
vulkan: preprocess mul_mat_id experts and discard workgroups more quickly (#18352)
jeffbolznv committed 31 days ago
Verified c9ced491
vulkan: optimize decodeFuncB in coopmat2 mul_mat_id shader (#18349)
jeffbolznv committed 31 days ago
Verified 7ac89021
vulkan: Use BK=32 for coopmat2 mul_mat_id (#18332)
jeffbolznv committed 31 days ago
Verified 9bf20d8a
vulkan: small dequantization improvements (#18380)
netrunnereve committed 31 days ago
Verified cb999704
vulkan: Support UPSCALE w/antialias (#18327)
jeffbolznv committed 31 days ago
Verified b96b82fc
vulkan: handle rope with large number of rows (#18306)
jeffbolznv committed 31 days ago
Verified 10dc500b
server : fix crash when seq_rm fails for hybrid/recurrent models (#18391)
o7si committed 32 days ago
Verified 4893cc07
docs: added note for pre SYCL Intel hardware (#18016)
alosslessdev committed 32 days ago
Verified af3be131
CANN: implement the SSM_CONV operator (#17737)
0Marble committed 32 days ago
Verified b07cda68
ggml-cuda: fix regex for arch list (#18371)
am17an committed 32 days ago
Verified 85c40c9b
cuda: optimize cumsum cub path (#18362)
am17an committed 32 days ago
Verified 83b3b1c2
ggml-cuda: fix blackwell native builds (#18361)
am17an committed 33 days ago
Verified b0fb0f0a
CANN: Add support for CONV_TRANSPOSE_1D when kernel size > 255 (#17934)
Intellouis committed 33 days ago
Verified e68c19b0
ggml : optimize cuda cumsum fallback kernel (#18343)
Aadeshveer committed 33 days ago
Verified c54bba86
server: (router) add stop-timeout option (#18350)
ngxson committed 33 days ago
Verified f5acfb2f
model: support MiMo-V2-Flash (#18328)
ngxson committed 33 days ago
Verified 4cbafad4
fit-params : fix race condition in fit-params output (#18276)
Aadeshveer committed 34 days ago
Verified c1842842
CUDA: experimental native mxfp4 support for blackwell (#17906)
am17an committed 34 days ago
Verified c8a2417d
model : support for LlamaBidirectionalModel architecture (#18220)
sfallah committed 34 days ago
Verified 54132f1b
vulkan: fix command buffer corruption in ggml_backend_vk_event_wait (#18302)
jeffbolznv committed 34 days ago
Verified 2a9ea202
CANN : refactor ACL graph cache (#17752)
wangweixuan committed 34 days ago
Verified ce7a6dc0
docs: Fix typos in SYCL documentation (#18269)
yoka committed 34 days ago
Verified 1ce0126b
vulkan: use fewer FA rows for small cache runs (#18280)
0cc4m committed 34 days ago
Verified 7f459c98
CANN: Uses yarn_ramp cache in ROPE (#17725)
TianHao324 committed 34 days ago
Verified cf2ffc02
common: add `LLAMA_ARG_OVERRIDE_TENSOR` env var for `-ot` arg (#18267)
ddh0 committed 34 days ago
Verified 10355dc7
server: return_progress to also report 0% processing state (#18305)
ngxson committed 34 days ago
Verified 5ee4e43f
webui: apply webui_settings on first load (#18223)
ServeurpersoCom committed 35 days ago
Verified 5b6c9bc0
Terms Of Use Privacy Policy Refund Policy Impressum