ggml-org/llama.cpp

Pull Requests Commits

metal : use FA-vec kernel up to batch size 20

ggerganov committed 119 days ago

Verified 78d70223

metal : optimize multi-sequence FA vec kernel

ggerganov committed 119 days ago

Verified fdfc7de7

batched-bench : fix pp batch contents

ggerganov committed 119 days ago

Verified f078c798

llama-bench : add defrag-thold, check for invalid ranges (#13487)

slaren committed 120 days ago

Verified cf0a43bb

opencl: remove unnecessary assert for `add` (#13257)

lhez committed 120 days ago

Verified f0d46ef1

clip : cap max image size 1024 for qwen vl model (#13478)

ngxson committed 120 days ago

Verified de4c07f9

llama/ggml: add LLM training support (#10544)

JohannesGaessler committed 120 days ago

Verified 10d2af0e

context : fix state io for memory-less contexts (#13470)

ggerganov committed 120 days ago

Verified 064cc596

server : allow content to be null in oaicompat_completion_params_parse (#13477)

anudit committed 120 days ago

Verified 91159ee9

llama-bench : accept ranges for integer parameters (#13410)

slaren committed 120 days ago

Verified 22cdab34

ggml-cpu: Integrate fp32=bf16xbf16 SME KleidiAI kernel (#13053)

eddnjjn committed 120 days ago

Verified a71a4075

CUDA: fix misaligned synchronization in FA (#13469)

JohannesGaessler committed 120 days ago

Verified 95e18884

ggml : add mrope kernel for metal (#13457)

ngxson committed 120 days ago

Verified df849192

enable dpcpp nightly builds with libraries (#13406)

AD2605 committed 120 days ago

Verified 14492144

mtmd : Use RMS norm for InternVL 3 38B and 78B mmproj (#13459)

city96 committed 121 days ago

Verified c1040239

tools : fix uninitialized llama_batch in server (#13436)

aumfer committed 121 days ago

Verified 9a390c48

scripts : exit compare-llama-bench.py gracefully when there's nothing to compare (#13451)

CISC committed 121 days ago

Verified 09232370

CUDA: fix crash with partial offloading of MoE (#13439)

JohannesGaessler committed 121 days ago

Verified 7474e00b

Add `--no-op-offload` to improve `-ot` pp perf in MoE models like llama4 400B (#13386)

hjc4869 committed 121 days ago

Verified 7f323a58

mtmd : support InternVL 3 38B and 78B mmproj (#13443)

city96 committed 121 days ago

Verified 3eac2093

mtmd : move helpers to dedicated file (#13442)

ngxson committed 121 days ago

Verified a634d75d

docs : Fix typo in InternVL3 model name (#13440)

99991 committed 122 days ago

Verified 62d4250e

CUDA: fix race conditions FlashAttention kernels (#13438)

JohannesGaessler committed 122 days ago

Verified 0208355f

vocab : add ByteDance-Seed/Seed-Coder (#13423)

CISC committed 122 days ago

Verified d2a4ef05

mtmd : add hard limit on image resolution for qwen2vl / qwen2.5vl (#13434)

ngxson committed 122 days ago

Verified 15e6125a

server : update docs (#13432)

ngxson committed 122 days ago

Verified 3b24d26c

llguidance : set tokenizer slices to default (#13424)

CISC committed 122 days ago

Verified 43dfd741

ci: free_disk_space flag enabled for intel variant (#13426)

Thammachart committed 122 days ago

Verified b064a51a

mtmd : support InternVL 2.5 and 3 (#13422)

ngxson committed 122 days ago

Verified 053367d1

CUDA: fix FlashAttention on Turing (#13415)

JohannesGaessler committed 122 days ago

Verified d8919424

Older