ggml-org/llama.cpp

Pull Requests Commits

convert : avoid dequantizing mxfp4 for GPT-OSS

compilade committed 98 days ago

d7f794ea

model-conversion : add trust_remote_code for orig model run [no ci] (#16751)

danbev committed 98 days ago

Verified 5a91109a

convert : handle pre-quantized models (#14810)

compilade committed 98 days ago

Verified f8f071fa

server: add memory breakdown print (#16740)

JohannesGaessler committed 98 days ago

Verified 0bf47a1d

convert : Make mistral-common dependency optional (#16738)

juliendenize committed 98 days ago

Verified dd62dcfa

mtmd-cli : allow using --jinja (#16718)

ngxson committed 99 days ago

Verified d0660f23

Manually link -lbsd to resolve flock symbol on AIX (#16610)

mehendarkarprajwal committed 99 days ago

Verified fe6a9882

ggml-cuda: use passed ops instead of hardcoded ops (#16712)

am17an committed 99 days ago

Verified 061f0eff

server : send partial stop string when <EOG> is reached (#15007)

matteoserva committed 99 days ago

Verified 8cf6b42d

sycl: use async memory allocation to fix crashes during graph recording (#16644)

mmichel11 committed 99 days ago

Verified 9de9672a

Add experimental ggml-hexagon backend for the Hexagon NPU (#16547)

max-krasnyansky committed 99 days ago

Verified 63d2fc46

Revert "ggml : Leverage the existing GGML_F32_VEC helpers to vectorize ggml_v…" (#16723)

slaren committed 99 days ago

Verified a2e0088d

webui: introduce OpenAI-compatible model selector in JSON payload (#16562)

ServeurpersoCom committed 99 days ago

Verified 9b9201f6

ggml : Leverage the existing GGML_F32_VEC helpers to vectorize ggml_vec_set_f32 for faster fills (#16522)

sirus20x6 committed 100 days ago

Verified 19a5a3ed

tests : fix test-thread-safety when compiling with multiple backends (#16699)

Acly committed 100 days ago

Verified d8eaa26e

CUDA: fix bug in topk-moe softmax (#16711)

am17an committed 100 days ago

Verified 9285325c

CUDA: topk-moe: add optional parameter for gpt-oss (#16649)

am17an committed 100 days ago

Verified 03792ad9

CUDA: better error for FA kernel with 0 occupancy (#16643)

JohannesGaessler committed 100 days ago

Verified 51d1a8c9

ggml: add ggml_can_fuse_subgraph (#16662)

am17an committed 101 days ago

Verified 4926419c

opencl: fix warnings and clean up profiling (#16688)

lhez committed 101 days ago

Verified 6ea37f57

vulkan: Handle FA with all -inf mask values (#16447)

jeffbolznv committed 101 days ago

Verified fb349848

sycl : add PAD_REFLECT_D1 operator support (#16145)

ye-NX committed 101 days ago

Verified 6de8ed75

model : add BailingMoeV2 support (#16063)

CISC committed 101 days ago

Verified 84bf3c67

Handle legacy 'context' attachments (#16687)

allozaur committed 101 days ago

Verified c9c1972e

ggml-alloc : fix leak when reusing a tensor with a larger size (#16679)

slaren committed 102 days ago

Verified b617cfd2

Prevent premature submission on IME input (#16673)

allozaur committed 102 days ago

Verified 79068501

Import/Export UX improvements (#16619)

allozaur committed 102 days ago

Verified 0e4a0cf2

Enable per-conversation loading states to allow having parallel conversations (#16327)

allozaur committed 102 days ago

Verified 13f2cfad

llama-batch: fix build fails with `-Werror=missing-braces` (#16614)

otegami committed 102 days ago

Verified 06332e28

readme: update bindings (#16651)

deadprogram committed 102 days ago

Verified 72d53e6c

Older