Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit history

Commits on Aug 9, 2026

• [CPU][MoE] Support DeepSeekV3 routing and group-size-32 INT4 checkpoints

  inisis

  committed
  a8cb9d0View commit details

  Copy full SHA for a8cb9d0

  View code at this pointBrowse repository at this point

• [KV Offload] Emit self-describing events for partial recurrent blocks (#51243)

  Show description for c423998
  

  chaunceyjiang

  and

  orozery

  authored
  c423998View commit details

  Copy full SHA for c423998

  View code at this pointBrowse repository at this point

• Bump Flashinfer version to 0.6.16.post3 (#50892)

  Show description for f18e10a
  

  3 peopleauthored
  f18e10aView commit details

  Copy full SHA for f18e10a

  View code at this pointBrowse repository at this point

• [Bugfix][KV Offload] Handle chunked local attention in offloading scheduler (#51161)

  Show description for eb24bc3
  

  3 peopleauthored
  eb24bc3View commit details

  Copy full SHA for eb24bc3

  View code at this pointBrowse repository at this point

• [KV Offload] Fix failed-load livelock by marking the lookup verdict as a miss (#49328)

  Show description for 1b0ce31
  RobbieJ

  authored
  1b0ce31View commit details

  Copy full SHA for 1b0ce31

  View code at this pointBrowse repository at this point

• [Migration] Migrate bitsandbytes support to OOT plugin (#43529)

  Show description for 073c510
  Isotr0py

  authored
  073c510View commit details

  Copy full SHA for 073c510

  View code at this pointBrowse repository at this point

• [V1][Metrics] Preserve prefix-cache stats on zero-output steps (#48668)

  Show description for dedbf6b
  

  3 peopleauthored
  dedbf6bView commit details

  Copy full SHA for dedbf6b

  View code at this pointBrowse repository at this point

• [Core] Make the GPU sync check thread-local and fix its suppressors (#51455)

  Show description for fbff187
  njhill

  authored
  fbff187View commit details

  Copy full SHA for fbff187

  View code at this pointBrowse repository at this point

• [Perf] Avoid some more unnecessary GPU<->CPU syncs (#51458)

  Show description for 9b0afeb
  njhill

  authored
  9b0afebView commit details

  Copy full SHA for 9b0afeb

  View code at this pointBrowse repository at this point

Commits on Aug 8, 2026

• [Bugfix][MRV2] Reserve spec-decode lookahead blocks in V2 warmup (#51438)

  Show description for d608dfa
  

  3 peopleauthored
  d608dfaView commit details

  Copy full SHA for d608dfa

  View code at this pointBrowse repository at this point

• [Perf][Sparse MLA] Drop the atomic contention in the index remap (#50365)

  Show description for 9e6be4a
  

  njhill

  and

  claude

  authored
  9e6be4aView commit details

  Copy full SHA for 9e6be4a

  View code at this pointBrowse repository at this point

• [Bugfix][Parser] Confirm reasoning end when an Inkling content block opens (#49876)

  Show description for 1c1077c
  Vegetog

  authored
  1c1077cView commit details

  Copy full SHA for 1c1077c

  View code at this pointBrowse repository at this point

• [Bugfix][Parser] Prevent Inkling block-end leakage with tools (#51391)

  Show description for 75231ef
  

  3 peopleauthored
  75231efView commit details

  Copy full SHA for 75231ef

  View code at this pointBrowse repository at this point

• [BugFix] Preserve divergent FA hits with external Mamba state (#51468)

  Show description for 58d3918
  

  4 peopleauthored
  58d3918View commit details

  Copy full SHA for 58d3918

  View code at this pointBrowse repository at this point

• [Bugfix] Fix LFM2 ShortConv prefix breaking quant ignore list (#51495)

  Show description for 0fe9a91
  

  xiaopusun

  and

  claude

  authored
  0fe9a91View commit details

  Copy full SHA for 0fe9a91

  View code at this pointBrowse repository at this point

• Add torch compile for qwen3_vl encoder (#40116)

  Show description for 653ebb5
  gty111

  authored
  653ebb5View commit details

  Copy full SHA for 653ebb5

  View code at this pointBrowse repository at this point

• [Bugfix][MM] Avoid device sync in FusedInputNorm initialization (#51435)

  Show description for 6e18901
  AndreasKaratzas

  authored
  6e18901View commit details

  Copy full SHA for 6e18901

  View code at this pointBrowse repository at this point

• [Bugfix][Multimodal] Fix PyNvVideoCodec video backend returning NCHW instead of NHWC (#51076)

  Show description for d3621c1
  davidjpyu

  authored
  d3621c1View commit details

  Copy full SHA for d3621c1

  View code at this pointBrowse repository at this point

• [BugFix] Reject invalid data-parallel RPC ports (#51469)

  Show description for 22a1759
  

  aoshen02

  and

  codex

  authored
  22a1759View commit details

  Copy full SHA for 22a1759

  View code at this pointBrowse repository at this point

• [Bugfix][Quantization] Fix INT8 W8A8 MoE crash in TritonExperts (#51411)

  Show description for 5eaa70d
  djramic

  authored
  5eaa70dView commit details

  Copy full SHA for 5eaa70d

  View code at this pointBrowse repository at this point

• [Bugfix] Fix ZMQ port TOCTOU race in shm_broadcast MessageQueue (#50960)

  Show description for f7ef489
  

  3 peopleauthored
  f7ef489View commit details

  Copy full SHA for f7ef489

  View code at this pointBrowse repository at this point

• [Kimi][MM] disable kimi_vit's dynamic torch.compile for TPU (#51196)

  Show description for 7f58e82
  lk-chen

  authored
  7f58e82View commit details

  Copy full SHA for 7f58e82

  View code at this pointBrowse repository at this point

• [Bugfix] Close usage telemetry HTTP sessions (#51219)

  Show description for 1c008a3
  matteso1

  authored
  1c008a3View commit details

  Copy full SHA for 1c008a3

  View code at this pointBrowse repository at this point

Commits on Aug 7, 2026

• [Perf] Optimize DeepSeek V3.2 sequence parallelism (#51434)

  Show description for e644c8c
  WoosukKwon

  authored
  e644c8cView commit details

  Copy full SHA for e644c8c

  View code at this pointBrowse repository at this point

• [Perf] Improve `--linear-backend` filtering (#48735)

  Show description for a801e71
  

  askliar

  and

  Aleksandr Skliar

  authored
  a801e71View commit details

  Copy full SHA for a801e71

  View code at this pointBrowse repository at this point

• [Perf] Narrow DeepSeek V3.2 eager CUDA graph region (#51425)

  Show description for 46b5864
  WoosukKwon

  authored
  46b5864View commit details

  Copy full SHA for 46b5864

  View code at this pointBrowse repository at this point

• [Bugfix] Drop stale layer kwarg from online MXFP4 kernel creation (fix precommit) (#51442)

  Show description for 9823714
  njhill

  authored
  9823714View commit details

  Copy full SHA for 9823714

  View code at this pointBrowse repository at this point

• [DSv32/GLM Perf] Skip short prefill topk for dense mha layer, 97.9% kernel level latency reduction (#51298)

  Show description for 4a11ed5
  yewentao256

  authored
  4a11ed5View commit details

  Copy full SHA for 4a11ed5

  View code at this pointBrowse repository at this point

• [Profiler] Stamp vLLM version/commit into torch profiler trace metadata (#51389)

  Show description for 99950b0
  elvircrn

  authored
  99950b0View commit details

  Copy full SHA for 99950b0

  View code at this pointBrowse repository at this point

• Add NVFP4 KV 4-over-6 scale search (#45187)

  Show description for 1c94e8d
  meenchen

  authored
  1c94e8dView commit details

  Copy full SHA for 1c94e8d

  View code at this pointBrowse repository at this point

• [Bugfix] when loading weights skip empty expert bias if model does not support them (#50937)

  Show description for 70456e5
  

  walterbm

  and

  hmellor

  authored
  70456e5View commit details

  Copy full SHA for 70456e5

  View code at this pointBrowse repository at this point

• [Frontend] Disable uvicorn signal handlers instead of racing them (#50916)

  Show description for ac70ce9
  njhill

  authored
  ac70ce9View commit details

  Copy full SHA for ac70ce9

  View code at this pointBrowse repository at this point

• [Online quantization] Add online MXFP4 quantization support (#49347)

  Show description for 3518110
  fxmarty-amd

  authored
  3518110View commit details

  Copy full SHA for 3518110

  View code at this pointBrowse repository at this point

• [MRv2 Feature] MR v2 weight offloading support (#51413)

  Show description for a671679
  yewentao256

  authored
  a671679View commit details

  Copy full SHA for a671679

  View code at this pointBrowse repository at this point

• [Perf] Raise Blackwell CUDA graph capture default to 1024 (#49390)

  LucasWilkinson

  authored
  021b7d9View commit details

  Copy full SHA for 021b7d9

  View code at this pointBrowse repository at this point