ggerganov/llama.cpp

Pull Requests Commits

ggml-quants : handle imatrix for MXFP4

compilade committed 59 days ago

d9b625ed

CANN: Add broadcast for softmax and FA (#15208)

hipudding committed 59 days ago

Verified be48528b

mtmd : Fix MinicpmV model converter and clip to avoid using hardcode. (#14750)

gryffindor-rr committed 59 days ago

Verified cf9e5648

chat : hotfix gpt-oss jinja raising an exception (#15243)

ngxson committed 59 days ago

Verified fba5c0d6

server : allow specifying reasoning_format in HTTP request (#15238)

ngxson committed 59 days ago

Verified 53d0a126

readme : update infra list (#15234)

malzag committed 59 days ago

Verified 27093afe

kv-cache : fix seq_rm with seq_id == -1 (#15226)

ggerganov committed 59 days ago

Verified 228f724d

kv-cache : log (debug) all streams in find_slot (#15176)

danbev committed 59 days ago

Verified cd3069df

convert : fix merge conflicts (#15229)

CISC committed 59 days ago

Verified 50e81bdf

perplexity : update comments/error msg to use decode [no ci] (#15227)

danbev committed 59 days ago

Verified 1ebbaddf

convert : improve Mistral models integration (#14737)

juliendenize committed 59 days ago

Verified a3a78742

kleidiai: fix unsigned overflow bug (#15150)

chaxu01 committed 59 days ago

Verified 002cb1bb

cuda: refactored ssm_scan and use CUB (#13291)

Your-Cheese committed 61 days ago

Verified 79c1160b

CUDA: add attention sinks for tile and wmma (#15178)

am17an committed 61 days ago

Verified 34c9d765

gguf-py : add Numpy MXFP4 de/quantization support (#15111)

compilade committed 62 days ago

Verified e54d41be

server-bench: external OAI servers, sqlite (#15179)

JohannesGaessler committed 62 days ago

Verified 4850b52a

ggml : fix field name when new ggml_backend (#14944)

aisk committed 62 days ago

Verified cd6983d5

vendor: sync minja (#15161)

ochafik committed 62 days ago

Verified 6c7e9a54

CUDA: attention sinks for mma FlashAttention (#15157)

JohannesGaessler committed 62 days ago

Verified 1425f587

opencl: support sink in `soft_max` (attn sinks) (#15152)

lhez committed 63 days ago

Verified aaa3d07a

convert : support non-mxfp4 HF model (#15153)

ngxson committed 63 days ago

Verified 50aa9389

vulkan: support fattn sinks (#15126)

jeffbolznv committed 63 days ago

Verified c4f53563

vulkan: Add env var to disable host visible vidmem (#15109)

jeffbolznv committed 63 days ago

Verified a0552c8b

llama : Support intern-s1 (#14875)

RunningLeon committed 63 days ago

Verified 99acbc99

HIP: add cmake option to enable compiler output of kernel resource usage metrics (#15103)

IMbackK committed 63 days ago

Verified 7ad67ba9

ggml: Skip backend library linking code when GGML_BACKEND_DL=ON (#15094)

ckastner committed 63 days ago

Verified 9a963895

CUDA: GEMM for FP32/FP16/BF16 and ne11 <= 16 (#15131)

JohannesGaessler committed 63 days ago

Verified 1d72c841

scripts: fix crash when --tool is not set (#15133)

JohannesGaessler committed 63 days ago

Verified 20638e4f

requirements : fix PyTorch uint64 compatibility (#15134)

danbev committed 64 days ago

Verified 36d3f00e

ggml: Add basic SET_ROWS support in WebGPU (#15137)

reeselevine committed 64 days ago

Verified 5fd160bb

Older