ggerganov/llama.cpp

Pull Requests Commits

model : avoid ggml_cont_3d for fused QKV weights

ggerganov committed 6 days ago

Verified f15d515e

CUDA: fuse adds, fuse add with rms norm (#15631)

am17an committed 6 days ago

Verified 009b709d

nvidia nemotron nano v2 (nemotronh) (#15507)

gabe-l-hart committed 6 days ago

Verified e8d99dd0

fix: Compute the full sum in llama-eval-callback, not just the sum of printed values (#15637)

gabe-l-hart committed 6 days ago

Verified a8bca68f

CUDA: add conv2d (#15635)

mnehete32 committed 7 days ago

Verified c97dc093

ggml-cpu: fix invalid hsum build in debug s390x (#15634)

taronaeo committed 7 days ago

Verified 6c442f42

ggml : fix SSM_SCAN for n_groups > 1 (#15625)

compilade committed 7 days ago

Verified 73804145

kv-cache : fix find_slot to not search for continuous slot (#15638)

ggerganov committed 7 days ago

Verified c8d0d14e

model : jina-embeddings-v3 support (#13693)

CISC committed 7 days ago

Verified 84ab83cc

scripts: add sqlite3 check for compare-commits.sh (#15633)

am17an committed 7 days ago

Verified 55042b36

kv-cache : remove LLAMA_SET_ROWS checks (#15505)

ggerganov committed 7 days ago

Verified 8a4280ce

gguf-py: byteswapping improvements (#12851)

AlekseiNikiforovIBM committed 7 days ago

Verified 64387f6e

cli : change log to warning to explain reason for stopping (#15604)

jrincayc committed 7 days ago

Verified d35a1e8c

model-conversion : add mmproj conversion target (#15628)

danbev committed 7 days ago

Verified 46d9caa2

cuda: Add cublasLt_static linking when GGML_STATIC is enabled (#15622)

matiaslin committed 7 days ago

Verified 5a0e3ef6

server: higher timeout for tests (#15621)

JohannesGaessler committed 8 days ago

Verified fbef0fad

presets : add qwen3-30B-a3b FIM (#15616)

ggerganov committed 8 days ago

Verified da54f9f1

HIP: Enable support for ggml_backend_cuda_register_host_buffer (#15615)

IMbackK committed 8 days ago

Verified 47373271

kv-cache : better estimate of n_kv for multi-sequence batches (#15610)

ggerganov committed 8 days ago

Verified 1bded5a3

CANN: refactor mask handling and improve performance in FA (#15561)

noemotiovon committed 8 days ago

Verified 1e748974

ggml-cpu : add basic RVV support for vector f32 ops (#15057)

xctan committed 8 days ago

Verified 1cf123a3

common : add -m to bash completion for --model [no ci] (#15591)

danbev committed 8 days ago

Verified fcca2182

OpenCL: add fused group_norm/norm, mul, add (#15314)

rmatif committed 8 days ago

Verified 86076f92

tests : fix test-opt with GGML_BACKEND_DL (#15599)

slaren committed 8 days ago

Verified bcbddcd5

SYCL: fix rms_norm_mul_add for tensor dim not a multiple of sg_size (#15592)

qnixsynapse committed 9 days ago

Verified 8b696861

mtmd : fix mtmd ios build (#15579)

fidoriel committed 9 days ago

Verified 8ce3ff1d

tests: add performance test for mul mat id (#15543)

netrunnereve committed 9 days ago

Verified 44b1efa4

llamafile: PowerPC Sgemm Optimization (#15558)

shalinib-ibm committed 9 days ago

Verified a6a58d64

graph : fix assert in memory-less build_attn (#15590)

ggerganov committed 9 days ago

Verified 0373486d

model-conversion : add qat-q4 quantization targets (#15588)

danbev committed 9 days ago

Verified 62cef26a

Older