ggerganov/llama.cpp

Pull Requests Commits

ggml-metal: FA tensor path: gate to the (256, 256) shape

ggerganov committed 28 days ago

60a1474a

ggml-metal: FA tensor kernel: register-light softmax in the per-d-block path

ggerganov committed 28 days ago

3f56636c

ggml-metal: FA tensor kernel: 32-wide chunks, register-light global-max softmax

ggerganov committed 28 days ago

c3148ebe

ggml-metal: hoist QK^T + softmax out of the d-block loop in the FA tensor kernel

ggerganov committed 28 days ago

36591b9d

ggml-metal: flash attention kernel for the MPP tensor API

ggerganov committed 28 days ago

21eb1b70

ggml-meta: propagate buffer usage and call init on the new tensors (#27586)

max-krasnyansky committed 28 days ago

Verified 5d5cb4c3

kleidiai: Rework KleidiAI Build System/Integration (#26077)

JonathanC-ARM committed 29 days ago

Verified d222767c

ci : update OS used for ROCM to Ubuntu 24.04 (#27681)

superm1 committed 29 days ago

Verified eab8ee41

rpc: support apple RDMA as an RPC transport (#26421)

ryan5rdx committed 29 days ago

Verified b114b473

devops: use GGML_NATIVE=OFF for OpenVINO (#27338)

krzsztf committed 29 days ago

Verified 0a5ac49b

server : reject prefilled assistant messages with tool calls (#27626)

kyo-zzz committed 29 days ago

Verified 1729ed53

chat : scope qwen3-coder workarounds (#27679)

aldehir committed 29 days ago

Verified 0cc5b149

ci : store ccache on HF buckets (test with cuda-ubuntu for now) (#27699)

CISC committed 29 days ago

Verified 790b5713

ui: ESLint config updates (#27700)

allozaur committed 29 days ago

Verified f1357e49

metal : null-check buffer alloc to fix OOM crash (#25371)

ykhrustalev committed 29 days ago

Verified 3737e413

llama.cpp : bump version to 0.3.0 (#27696)

ggerganov committed 29 days ago

Verified c1d0e7a0

ggerganov committed 29 days ago

81191aff

ggml : bump version to 0.22.0 (ggml/1607)

ggerganov committed 29 days ago

93882361

grammar : parse \- in char classes as literal hyphen (#27591)

NIXKnight committed 29 days ago

Verified eb25b726

sycl : mark tq2_0 as not supported (#27660)

arthw committed 29 days ago

Verified 814d84bc

webgpu : fix handling of infinity values during ARGSORT and TOP_K (#27538)

fairydreaming committed 29 days ago

Verified 5ea87dda

metal : per-device tuned (Q, NE) for flash-attn vec (#26570)

forforever73 committed 30 days ago

Verified f280b269

metal: per-op source split + parallel compile (#26561)

forforever73 committed 30 days ago

Verified b615f5b4

misc : read repetition_penalty from generation_config.json (#27659)

tdakhran committed 30 days ago

Verified b3c3b96a

tests : disable DOTS3NOTE arch test for WebGPU (#27654)

fairydreaming committed 30 days ago

Verified 75844307

convert: fix GLM regression in index_tensors (#27655)

jacekpoplawski committed 30 days ago

Verified 71cc86fa

ggml : shorten virtual device naming in CUDA and Metal (#27608)

ggerganov committed 30 days ago

Verified a14dba68

webgpu : reorder includes since V that appears in common_decls.tmpl may be defined as K in flash_attn_decls.tmpl if KV_OVERLAP (#27545)

fairydreaming committed 30 days ago

Verified c1c766da

mtmd: video: fix moov atom at the end of file (#27596)

ngxson committed 30 days ago

Verified 160c6b0b

ci : apply ccache-clear with older/min/dry-run to all ccache jobs (#27602)

ggerganov committed 30 days ago

Verified 985b1491

Older