Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Mar 1, 2026

• Revert "[Bugfix] Disable TRTLLM attention with KV transfer enabled (#33192)" (#34832)

  Show description for 57a96e2
  ZhanqiuHu

  authored
  57a96e2

  Copy full SHA for 57a96e2

• [torch.compile] Undo the fast_moe_cold_start hack in torch>=2.11 (#35475)

  Show description for e82fbee
  zou3519

  authored
  e82fbee

  Copy full SHA for e82fbee

• [Bugfix] Fix dtype mismatch in RMSNormGated.forward_native() during torch.compile (#35256)

  Show description for 6290470
  haosdent

  authored
  6290470

  Copy full SHA for 6290470

• [Model Runner V2] Use block table apis for capture inputs (#35671)

  Show description for 72f4d16
  WoosukKwon

  authored
  72f4d16

  Copy full SHA for 72f4d16

• fix(mxfp4): return is_monolithic=False when LoRA is enabled for Triton backend (#35382)

  Show description for 5a43550
  yoonsnowdev

  authored
  5a43550

  Copy full SHA for 5a43550

• [MISC] Fixing a null reference by removing parallel_utils from mypy EXCLUDE (#35630)

  Show description for 59d7af9
  taneem-ibrahim

  authored
  59d7af9

  Copy full SHA for 59d7af9

• [Mamba1] - Kernel Level Chunk Alignment for Prefix Caching (#34798)

  Show description for bbf81f9
  Josephasafg

  authored
  bbf81f9

  Copy full SHA for bbf81f9

• [Model Runner V2] Minor refactoring for EncoderRunner (#35628)

  Show description for da543d1
  WoosukKwon

  authored
  da543d1

  Copy full SHA for da543d1

• [AMD][CI] Support Triton attention with ExampleConnector (#34931)

  Show description for 87d319c
  rjrock

  authored
  87d319c

  Copy full SHA for 87d319c

• Fix typo: implictly -> implicitly in isaac.py docstring (#35646)

  lin-shh

  authored
  a9ec392

  Copy full SHA for a9ec392

• [Bugfix][Model] Fix Qwen3.5/Qwen3Next ignoring --dtype flag on older GPUs (#35617)

  lailoo

  authored
  afd089f

  Copy full SHA for afd089f

• Add TMA support to fused_moe_lora kernel (#32195)

  Show description for 3ecd0bf
  

  gnovack

  and

  jeejeelee

  authored
  3ecd0bf

  Copy full SHA for 3ecd0bf

Commits on Feb 28, 2026

• [Model Runner V2] Add ModelStateInterface [4/N] (#35621)

  Show description for e3eb146
  WoosukKwon

  authored
  e3eb146

  Copy full SHA for e3eb146

• [Bugfix] Fix Anthropic API base64 image handling in Messages endpoint (#35557)

  Show description for 95a395d
  voipmonitor

  authored
  95a395d

  Copy full SHA for 95a395d

• [Chore] Cleanup BNB utilization dead code (#35620)

  Show description for e94b263
  Isotr0py

  authored
  e94b263

  Copy full SHA for e94b263

• [Deprecation] Deprecate code in 0.17 as scheduled (#35441)

  Show description for e113a30
  

  yewentao256

  and

  DarkLight1337

  authored
  e113a30

  Copy full SHA for e113a30

• [Benchmark] Avoid unnecessary video download in MMVU (#35618)

  Show description for 1dafb29
  DarkLight1337

  authored
  1dafb29

  Copy full SHA for 1dafb29

• [Fix] Avoid sending image input to other PP ranks (#35405)

  Show description for 49b9ae3
  

  emricksini-h

  and

  ywang96

  authored
  49b9ae3

  Copy full SHA for 49b9ae3

• Fix Qwen3_5MTP packed_modules_mapping for gate_up_proj (#35581)

  cwazai

  authored
  63d7972

  Copy full SHA for 63d7972

• custom dataset img support base64 (#35280)

  Show description for c68e69f
  flutist

  authored
  c68e69f

  Copy full SHA for c68e69f

• [Feat] Add CUDA torch fallbacks for fp8_mqa_logits/fp8_paged_mqa_logits_torch function (#35271)

  Show description for 7e08c22
  chaunceyjiang

  authored
  7e08c22

  Copy full SHA for 7e08c22

• add io_process_plugin for sparse embedding (#34214)

  Show description for 8e75d88
  

  staugust

  and

  DarkLight1337

  authored
  8e75d88

  Copy full SHA for 8e75d88

• [Feature]Supports Anthropic Thinking Block (#33671)

  Show description for 0892d1a
  

  3 peopleauthored
  0892d1a

  Copy full SHA for 0892d1a

• Add padding support to wvSplitK solution for skinny GEMMs (#33762)

  Show description for 7600642
  amd-hhashemi

  authored
  7600642

  Copy full SHA for 7600642

• [ROCm][CI] Parametrize vision score tests across attention backends with per-backend tolerances (#35571)

  Show description for 1e69c04
  AndreasKaratzas

  authored
  1e69c04

  Copy full SHA for 1e69c04

• [Benchmark] Improve UX of sweep scripts (#35600)

  Show description for 4292e3b
  DarkLight1337

  authored
  4292e3b

  Copy full SHA for 4292e3b

• [Benchmark] Rename SLA Finder to Workload Explorer (#35586)

  Show description for 24d6ea8
  DarkLight1337

  authored
  24d6ea8

  Copy full SHA for 24d6ea8

• [Misc] Change logging level from info to debug for tool parser import (#35575)

  Show description for 57c86c0
  chaunceyjiang

  authored
  57c86c0

  Copy full SHA for 57c86c0

• [CI] add trainer_send_weights for MockWeightTransferEngine (#35589)

  Show description for 06254d4
  chaunceyjiang

  authored
  06254d4

  Copy full SHA for 06254d4

• [ROCm][CI] Expose tests to AMD production CI and fix amdsmi heap corruption (#35071)

  Show description for f5d1281
  AndreasKaratzas

  authored
  f5d1281

  Copy full SHA for f5d1281

• [ROCm] Derive device capability from GCN arch string without CUDA init (#35069)

  Show description for 94029ff
  AndreasKaratzas

  authored
  94029ff

  Copy full SHA for 94029ff

• [ROCm][CI] Adding infiniband mappings for moriio tests (#35170)

  Show description for 88e8525
  AndreasKaratzas

  authored
  88e8525

  Copy full SHA for 88e8525

• [EPLB] Enforce sync eplb for NCCL-based all2all backend (#35212)

  Show description for b2d8b42
  ilmarkov

  authored
  b2d8b42

  Copy full SHA for b2d8b42

• [Bugfix] Move chat completion response_format validation to Pydantic model_validator (#35510)

  Show description for 1d5ab5d
  umut-polat

  authored
  1d5ab5d

  Copy full SHA for 1d5ab5d

• [Bugfix] Propagate compilation_time from workers to main process for TP>1 (#35503)

  Show description for 7b346ba
  huydhn

  authored
  7b346ba

  Copy full SHA for 7b346ba