ngxson/llama.cpp

Pull Requests Commits
cuda: fix race condition in cumsum (#18448)
am17an committed 5 days ago
Verified 5fa66c6e
ci : re-enable rocm build on amd64 (#18439)
TimNN committed 5 days ago
Verified 382808c1
HIP: Use mmq on MFMA devices for MUL_MAT_ID in cases where a lot of splits would be generated (#18202)
IMbackK committed 5 days ago
Verified 4ffc47cb
model : Plamo3 support (#17304)
mmnga committed 5 days ago
Verified 9c675c71
Revert "ggml-cuda: use CMAKE_CUDA_ARCHITECTURES if set when GGML_NATIVE=ON (#18413)" (#18426)
am17an committed 5 days ago
Verified 07a0c4ba
rpc: fix segfault on invalid endpoint format (#18387)
o7si committed 6 days ago
Verified 60f17f56
llama-fit-params: fix step size for last device (#18415)
JohannesGaessler committed 6 days ago
Verified f8d561eb
github: update issue templates [no ci] (#18410)
JohannesGaessler committed 6 days ago
Verified e59efe6a
mtmd: clarify that we no longer accept AI-generated PRs (#18406)
ngxson committed 6 days ago
Verified cffa5c46
cmake: Added more x86_64 CPU backends when building with `GGML_CPU_ALL_VARIANTS=On` (#18186)
bberberov committed 6 days ago
Verified 94de74e7
ggml-cuda: use CMAKE_CUDA_ARCHITECTURES if set when GGML_NATIVE=ON (#18413)
QDelta committed 6 days ago
Verified 4fd59e84
opencl: allow resizing transpose buffers (#18384)
lhez committed 6 days ago
Verified 08566977
llama-fit-params: fix overflow check (#18354)
JohannesGaessler committed 6 days ago
Verified a4bf3588
llama: fix magic number of 999 for GPU layers (#18266)
JohannesGaessler committed 6 days ago
Verified 026d2ad4
ggml-cuda: Use same regex for GGML_NATIVE=OFF (#18407)
am17an committed 7 days ago
Verified 06705fdc
llama_fit_params: return enum for fail vs. error (#18374)
JohannesGaessler committed 7 days ago
Verified a52dc60b
llama-fit-params: fix Gemma 3 calculation (#18372)
JohannesGaessler committed 7 days ago
Verified 9045c9af
vulkan: preprocess mul_mat_id experts and discard workgroups more quickly (#18352)
jeffbolznv committed 7 days ago
Verified c9ced491
vulkan: optimize decodeFuncB in coopmat2 mul_mat_id shader (#18349)
jeffbolznv committed 7 days ago
Verified 7ac89021
vulkan: Use BK=32 for coopmat2 mul_mat_id (#18332)
jeffbolznv committed 7 days ago
Verified 9bf20d8a
vulkan: small dequantization improvements (#18380)
netrunnereve committed 7 days ago
Verified cb999704
vulkan: Support UPSCALE w/antialias (#18327)
jeffbolznv committed 7 days ago
Verified b96b82fc
vulkan: handle rope with large number of rows (#18306)
jeffbolznv committed 7 days ago
Verified 10dc500b
server : fix crash when seq_rm fails for hybrid/recurrent models (#18391)
o7si committed 7 days ago
Verified 4893cc07
docs: added note for pre SYCL Intel hardware (#18016)
alosslessdev committed 8 days ago
Verified af3be131
CANN: implement the SSM_CONV operator (#17737)
0Marble committed 8 days ago
Verified b07cda68
ggml-cuda: fix regex for arch list (#18371)
am17an committed 8 days ago
Verified 85c40c9b
cuda: optimize cumsum cub path (#18362)
am17an committed 8 days ago
Verified 83b3b1c2
ggml-cuda: fix blackwell native builds (#18361)
am17an committed 8 days ago
Verified b0fb0f0a
CANN: Add support for CONV_TRANSPOSE_1D when kernel size > 255 (#17934)
Intellouis committed 9 days ago
Verified e68c19b0
Terms Of Use Privacy Policy Refund Policy Impressum