ggerganov/llama.cpp

Pull Requests Commits

ui : model memory-fit estimation

allozaur committed 10 days ago

3e43ff1b

ui : strip provider tilde prefix from hub avatar urls

allozaur committed 10 days ago

c9462ce7

ui : Hugging Face Hub data layer

allozaur committed 10 days ago

f9f41f64

ui : model id grammar for sidecars, quants and capability parsing

allozaur committed 10 days ago

0bdbf547

ui : type-safe API types, fetch helpers and download-ready models store plumbing

allozaur committed 10 days ago

e5278550

server : fix deadlock when removing a finished download

allozaur committed 10 days ago

5963c000

common : resolve <quant>-<sidecar> download tags and list cached sidecars

allozaur committed 10 days ago

0949acee

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

y198nt committed 10 days ago

Verified 60199339

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

gaugarg-nv committed 10 days ago

Verified b04d4e56

qwen4exp: add hc ops (#28901)

am17an committed 10 days ago

Verified 37b53fd4

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

WenqiangJia2026 committed 10 days ago

Verified fccf7166

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

aldehir committed 10 days ago

Verified 0bec16e3

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

SG-Amadeus committed 10 days ago

Verified d4365d95

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

mdegans committed 10 days ago

Verified 0a8b29a6

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

CISC committed 10 days ago

Verified 583926e3

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

asbelin committed 10 days ago

Verified e13469a3

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

jhen0409 committed 10 days ago

Verified 930e2fa5

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

trivikram-reddy1 committed 10 days ago

Verified 72b590d6

HIP: Enable AllReduce for ROCm (#27825)

Stastez committed 11 days ago

Verified 38a5b42d

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

wanghqc committed 11 days ago

Verified 9f31776c

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

am17an committed 11 days ago

Verified d1d3c339

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

JohannesGaessler committed 11 days ago

Verified 6011c34c

rpc : hash-cache only weights (#28789)

Patt92 committed 11 days ago

Verified 76098465

cuda: support row-contiguous SUM_ROWS (#26308)

MohamedElashri committed 11 days ago

Verified 54315813

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

cpeterso committed 11 days ago

Verified 9e717162

vulkan: support sparse Flash Attention (#28105)

0cc4m committed 11 days ago

Verified fc82583e

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

wine99 committed 11 days ago

Verified 77d554b2

opencl: add generic ssm_scan (#28881)

lhez committed 11 days ago

Verified 6ec1a7e9

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

taronaeo committed 11 days ago

Verified 1af6c65d

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

wyanzhao committed 11 days ago

Verified 1e7bcf3d

Older