ngxson/llama.cpp

Pull Requests Commits

cuda: fuse MoE weighted expert reduction (#25952)

anujj committed 1 hour ago

Verified 3466812d

kv-cells: look up the n-gram history in the sequence position index (#28040)

ServeurpersoCom committed 3 hours ago

Verified b356fa26

context : autoscale n_ctx_train when yarn scaling specified (#28030)

CISC committed 4 hours ago

Verified dfc29b64

models : appropriately flag noscan ssm_a tensors (#28121)

CISC committed 4 hours ago

Verified f28493c7

model : fix gemma4-assistant (#28183)

CISC committed 4 hours ago

Verified 73159c30

model : load relevant arrays with n_layer_all (#28173)

CISC committed 4 hours ago

Verified d11b3cc7

Revert "sycl : add Kronecker product FWHT support for sizes 384, 640, 768, 12…" (#28184)

Titaniumtown committed 5 hours ago

Verified c845263f

sycl : add Kronecker product FWHT support for sizes 384, 640, 768, 1280 (#28016)

philip-jingxin committed 6 hours ago

Verified 1f3d3187

metal : add fa-vec tuning for M2 Pro (#28122)

lstolcman committed 8 hours ago

Verified 8887a48f

metal : add fa-vec tunings for A18 Pro (MacBook Neo) (#28152)

jhen0409 committed 8 hours ago

Verified be789c34

model : load hparams.n_layer_nextn before n_layer() calls (#28159)

CISC committed 9 hours ago

Verified 9d817213

metal : fix more leaks due to missing autoreleasepools (#27883)

nikwen committed 9 hours ago

Verified fe2120bc

metal : add fa-vec tuning for M2 Max (#28015)

ggerganov committed 11 hours ago

Verified d08c7872

sycl : support limit max alloc memory within 2GB for host-pinned memory (#27559)

arthw committed 11 hours ago

Verified 5eec3ad0

qwen4exp: fix seq_cp, block position keying, mtmd input, cuda abort, add tests (#27941)

danielhanchen committed 11 hours ago

Verified 36b10154

tests : fix log verbosity for test-llama-archs (#28147)

ggerganov committed 11 hours ago

Verified d086dbb3

quantize: row-slab stream to avoid thread starvation (#27830)

ngxson committed 12 hours ago

Verified 1b89a43e

metal: enable Metal 4.0 tensor API on M5+/A19+ (#27461)

JamesFranc committed 12 hours ago

Verified d5d993a0

ci: Bump ggml-org/ccache-action to v1.2.24 (#28083)

luhenry committed 12 hours ago

Verified 234a6eba

kleidiai : Update KleidiAI Documentation (#26078)

JonathanC-ARM committed 13 hours ago

Verified 518b7623

qwen4exp: support recurrent state rollback (#28123)

ServeurpersoCom committed 17 hours ago

Verified 0eadefeb

qwen4exp: sum the indexer heads by slices (#28023)

ServeurpersoCom committed 17 hours ago

Verified 09412af3

metal : add fa-vec tunings for M1 Ultra (#28088)

ozgursoy committed 23 hours ago

Verified 458681e1

CUDA: XOR swizzle flash attn K,V smem fp16 tiles (#25635)

ynankani committed 1 day ago

Verified e4b9af00

metal : add concat support for quantized types (#28116)

ggerganov committed 1 day ago

Verified ab0b3bd3

AVX2: Speed up large batch size prompt processing of IQ models (#27402)

bartowski1182 committed 1 day ago

Verified 85c55223

metal : add top-k radix implementation (#28073)

ggerganov committed 1 day ago

Verified 2a74817f

kv-cache : optimize restoring non-contiguous cells (#27991)

itsnotoger committed 1 day ago

Verified 2d8d612e

opencl: tune the quant paths for Intel Xe-LP GPUs to improve its TG and PP performance (#26438)

wanghqc committed 1 day ago

Verified 010be968

ui: copy the displayed text of grouped agentic responses (#27832)

ServeurpersoCom committed 1 day ago

Verified 774ee0e2

Older