ggml-org/llama.cpp

Pull Requests Commits

json : support `enum` values within `allOf` (#15830)

aldehir committed 2 hours ago

Verified 7057faf6

media : add llama1 icon (#15878)

06kellyjac committed 4 hours ago

Verified fe1c92cd

vulkan: sort graph to allow more parallel execution (#15850)

jeffbolznv committed 5 hours ago

Verified e68aa10d

CUDA: generate_cu_files.py - add missing mxfp4 (#15880)

am17an committed 6 hours ago

Verified 0a16bf52

chat : Deepseek V3.1 reasoning and tool calling support (OpenAI Style) (#15533)

createthis committed 8 hours ago

Verified 88021565

server : bring back timings_per_token (#15879)

ngxson committed 8 hours ago

Verified 56920f56

cuda : fix supports_op condition for get_rows when number of blocks is too large (#15868)

ggerganov committed 12 hours ago

Verified b0d52998

metal : refactor + optimize (#15857)

ggerganov committed 13 hours ago

Verified f28d4f4a

ggml: allow casting between f32 and i32 (#15783)

ngxson committed 13 hours ago

Verified 9fcb29f2

CUDA: non-contiguous src0 not supported for PAD (#15869)

CISC committed 13 hours ago

Verified 5ef22d28

convert : force setting sliding_window from original config (#15867)

danbev committed 16 hours ago

Verified 233d773d

batched-bench : fix llama_synchronize usage during prompt processing (#15835)

ggerganov committed 16 hours ago

Verified a885dcff

context : fix n_outputs during reserve (#15858)

ggerganov committed 16 hours ago

Verified 663027fd

model : avoid ggml_cont_3d for fused QKV weights (#15662)

ggerganov committed 16 hours ago

Verified cf0e3ba1

tests: large sizes for get_rows (#15687)

jeffbolznv committed 19 hours ago

Verified d413dca0

CANN: Stream sync between devices for acl_graph (#15809)

noemotiovon committed 21 hours ago

Verified 85ca66a7

vulkan: support im2col_3d (#15795)

jeffbolznv committed 1 day ago

Verified 3976dfbe

ggml-cpu: clean up s390x SIMD (#15855)

taronaeo committed 1 day ago

Verified d36e61c5

vulkan: Support pad_ext (#15794)

jeffbolznv committed 1 day ago

Verified c97b5e58

vulkan: Use larger loads in scalar/coopmat1 matmul (#15729)

jeffbolznv committed 1 day ago

Verified 267e9986

ggml WebGPU: remove userdata from request adapter callback (#15527)

danbev committed 1 day ago

Verified 3b15924d

CUDA: faster tile FA (Pascal/AMD), headsize 256 (#15769)

JohannesGaessler committed 2 days ago

Verified 79bc4292

kleidiai: generalize compute_forward_kv_cache to compute_forward_fp16 (#15817)

chaxu01 committed 2 days ago

Verified c4df49a4

server : speed up tests (#15836)

ngxson committed 2 days ago

Verified 3c3635d2

server : implement prompt processing progress report in stream mode (#15827)

ngxson committed 2 days ago

Verified 61bdfd52

ggml-cpu: document use of "free" memory [no ci] (#15834)

JohannesGaessler committed 2 days ago

Verified 01806e77

ggml-cpu: drop support for nnpa intrinsics (#15821)

taronaeo committed 2 days ago

Verified 186415d5

aLoRA Support (#15327)

gabe-l-hart committed 3 days ago

Verified fd621880

ci : exempt correct research label (#15825)

CISC committed 3 days ago

Verified 4281c7b3

Thinking model disabled assistant prefill (#15404)

gabe-l-hart committed 3 days ago

Verified 5fac79cb

Older