ggml-org/llama.cpp

Pull Requests Commits

ggerganov committed 23 days ago

d093cf70

hexagon: Support for K-Quants Q4_K and Q6_K (#28994)

rjtokenring committed 25 days ago

Verified 1ec81880

hexagon: accept the zeroed rope probe in supports_op (#28995)

rjtokenring committed 25 days ago

Verified 82324fc5

models : allow Nemotron-H models to only define layer_norm_epsilon (#28989)

frozenblade1224 committed 25 days ago

Verified 7ceed873

model : add support for HrmTextForCausalLM (DFM Mimir 1B) (#27625)

noctrex committed 25 days ago

Verified 7d6f5d02

CUDA/HIP: improve access patterns in im2col (#28013)

IMbackK committed 25 days ago

Verified 83078fec

spacemit : fix wrong transpose function for int16 data (#25161)

I3eg1nner committed 25 days ago

Verified f266648f

rpc : invalidate cached compute graph when a referenced buffer is freed (#24292)

y198nt committed 25 days ago

Verified 60199339

Change max context length for auto-fitting with unified KV (#28849)

gaugarg-nv committed 25 days ago

Verified b04d4e56

qwen4exp: add hc ops (#28901)

am17an committed 25 days ago

Verified 37b53fd4

HIP: broaden MoE ncols_opt tile heuristic on RDNA3.5 architecture (#28935)

WenqiangJia2026 committed 25 days ago

Verified fccf7166

chat : force `\n</think>` on reasoning budget end for qwen3-coder (#28869)

aldehir committed 25 days ago

Verified 0bec16e3

vulkan: make MUL_MAT_ID BN/2 tail unconditional (#28923)

SG-Amadeus committed 25 days ago

Verified d4365d95

metal: fix NaN in mul_mm_id when activations exceed f16 range (#26223)

mdegans committed 25 days ago

Verified 0a8b29a6

ci : add self-hosted webgpu to hf-jobs (#28712)

CISC committed 25 days ago

Verified 583926e3

llama-bench: support --version to print build info (#28971)

asbelin committed 25 days ago

Verified e13469a3

hexagon: add back missing contiguous fast-path and hvx_copy_uu for each run (#28886)

jhen0409 committed 26 days ago

Verified 930e2fa5

hex-cpy: use dma if src and dst are contiguous (#28906)

trivikram-reddy1 committed 26 days ago

Verified 72b590d6

HIP: Enable AllReduce for ROCm (#27825)

Stastez committed 26 days ago

Verified 38a5b42d

opencl: choose the MoE expert matmul by batch size for speculative decoding/MTP (#27637)

wanghqc committed 26 days ago

Verified 9f31776c

ci: build MUSA for only 1 arch (#28944)

am17an committed 26 days ago

Verified d1d3c339

docs: Rule of thumb for AI review time [no ci] (#28945)

JohannesGaessler committed 26 days ago

Verified 6011c34c

rpc : hash-cache only weights (#28789)

Patt92 committed 26 days ago

Verified 76098465

cuda: support row-contiguous SUM_ROWS (#26308)

MohamedElashri committed 26 days ago

Verified 54315813

models : move build_arch_graph() after graph() template specialization (#28934)

cpeterso committed 26 days ago

Verified 9e717162

vulkan: support sparse Flash Attention (#28105)

0cc4m committed 26 days ago

Verified fc82583e

OpenVINO: optimize stateful decode and GPU MoE inference (#28638)

wine99 committed 26 days ago

Verified 77d554b2

opencl: add generic ssm_scan (#28881)

lhez committed 26 days ago

Verified 6ec1a7e9

ci: bump kleidiai runners from 22.04 to 24.04 (#28885)

taronaeo committed 26 days ago

Verified 1af6c65d

metal : add FA kernels for HSK=96, HSV=64 (MiniCPM3) (#28599)

wyanzhao committed 26 days ago

Verified 1e7bcf3d

Older