Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Apr 20, 2026

• [Bugfix] Properly initialize `PerTensorScaleParameter` for fused-on-disk checkpoints (#39765)

  Show description for fcb31c1
  

  Alnusjaponica

  and

  gemini-code-assist[bot]

  authored
  fcb31c1

  Copy full SHA for fcb31c1

• [Doc] Fix typos in token_embed pooling documentation (#40266)

  Show description for d886c26
  YifanLi3

  authored
  d886c26

  Copy full SHA for d886c26

• [BugFix][XPU] fix lora ops bgmv_expand size not match (#39989)

  Show description for 898beca
  

  Liangliang-Ma

  and

  jikunshang

  authored
  898beca

  Copy full SHA for 898beca

Commits on Apr 19, 2026

• [ci] Make ecr authenticate non blocking (#40305)

  Show description for 629d45e
  khluu

  authored
  629d45e

  Copy full SHA for 629d45e

• [ROCm] Fix cu_seqlens_q off-by-one in AITER FA speculative decode path (#39120)

  Show description for f150107
  Bortlesboat

  authored
  f150107

  Copy full SHA for f150107

• Fix MoE backend selection for LoRA (unquantized MoE) (#40273)

  Show description for d1135a5
  danisereb

  authored
  d1135a5

  Copy full SHA for d1135a5

• Optimize nemotron VL image/video preprocessing (#40283)

  Show description for 982beae
  

  netanel-haber

  and

  milesial

  authored
  982beae

  Copy full SHA for 982beae

• [FEAT] [Perf] [Gemma4] Fused Gemma4 Routing Function Triton (#39083)

  Show description for 45232a4
  tjtanaa

  authored
  45232a4

  Copy full SHA for 45232a4

• [Bugfix] Kimi-K2 tool parser streaming - fix token leakage, argument truncation, and content dropping (#38579)

  Show description for 03ce1c6
  sfeng33

  authored
  03ce1c6

  Copy full SHA for 03ce1c6

• [KV Offload] Pass request context (#39185)

  Show description for 4353c9c
  omerpaz95

  authored
  4353c9c

  Copy full SHA for 4353c9c

• [KV Connector] Allow metrics of multiple connectors of same types in multi connector. (#40010)

  Show description for 4b7f5ea
  

  omerpaz95

  and

  orozery

  authored
  4b7f5ea

  Copy full SHA for 4b7f5ea

Commits on Apr 18, 2026

• [Frontend] Preserve structured output special tokens in offline LLM.chat (#39352)

  Show description for 38907e4
  

  lucianommartins

  and

  sfeng33

  authored
  38907e4

  Copy full SHA for 38907e4

• [Bugfix] Guard mxfp4_experts_quant bindings on ENABLE_NVFP4_SM100 (#40191)

  Show description for d0359f3
  

  3 peopleauthored
  d0359f3

  Copy full SHA for d0359f3

• [Attention] TurboQuant: remove redundant random signs, add prior art attribution (#40194)

  Show description for ed0622e
  dalistarh

  authored
  ed0622e

  Copy full SHA for ed0622e

• Added general ND x ND matmul and unit test for it (#39909)

  Show description for b5f6c5f
  YM2132

  authored
  b5f6c5f

  Copy full SHA for b5f6c5f

• [DOC] Add fuse_minimax_qk_norm (#39782)

  Show description for bfde49e
  jeejeelee

  authored
  bfde49e

  Copy full SHA for bfde49e

• [Refactor] Drop direct dependency on librosa (#39079)

  Show description for 153ba7f
  

  NickCao

  and

  claude

  authored
  153ba7f

  Copy full SHA for 153ba7f

• [ZenCPU] AMD Zen CPU Backend with supported dtypes via zentorch weekly (#39967)

  Show description for 87518c3
  Chinmay-Kulkarni-AMD

  authored
  87518c3

  Copy full SHA for 87518c3

• [Bugfix] Fix k_proj's bias for GLM-ASR (#40160)

  Show description for aeee7ef
  rishaps

  authored
  aeee7ef

  Copy full SHA for aeee7ef

• [Doc] Fix outdated source reference comment in anthropic/serving.py (#40189)

  Show description for cda19ec
  z1ying

  authored
  cda19ec

  Copy full SHA for cda19ec

• [Frontend] Add multimodal support to /inference/v1/generate endpoint (#38405)

  Show description for 80b1823
  

  nithinvc

  and

  DarkLight1337

  authored
  80b1823

  Copy full SHA for 80b1823

• [Doc] Add Realtime Transcription section to supported_models.md (#39845)

  Show description for d0697cc
  z1ying

  authored
  d0697cc

  Copy full SHA for d0697cc

• [Core] Reduce mm scheduler, get_num_embed overhead (#40143)

  Show description for b075552
  

  milesial

  and

  gemini-code-assist[bot]

  authored
  b075552

  Copy full SHA for b075552

• [XPU] fix all_reduce all-zero accuracy issue under torch.compile (#39844)

  Show description for 993859c
  

  chaojun-zhang

  and

  mergify[bot]

  authored
  993859c

  Copy full SHA for 993859c

• [CI] Speed up test_fused_marlin_moe (#40178)

  Show description for 48a65cc
  mgoin

  authored
  48a65cc

  Copy full SHA for 48a65cc

• [XPU]fake impl for xpu fp8_gemm (#39984)

  Show description for 55842a8
  

  xinyu-intel

  and

  jikunshang

  authored
  55842a8

  Copy full SHA for 55842a8

Commits on Apr 17, 2026

• Remove outdated tests test_mixtral_moe and test_duplicated_ignored_sequence_group (#40175)

  Show description for 1f45e83
  mgoin

  authored
  1f45e83

  Copy full SHA for 1f45e83

• [Kernel] Add MXFP4 W4A4 CUTLASS MoE kernel for SM100 (#37463)

  Show description for a8bffaa
  mgoin

  authored
  a8bffaa

  Copy full SHA for a8bffaa

• [Kernel] [Helion] Force disable HOP path due to performance regression (#40171)

  Show description for 5cddddd
  

  gmagogsfm

  and

  claude

  authored
  5cddddd

  Copy full SHA for 5cddddd

• [ROCm] Fix TurboQuant on ROCm: backend routing, flash-attn compat, int64 overflow (#39953)

  Show description for 6ef1efd
  aditi-amd

  authored
  6ef1efd

  Copy full SHA for 6ef1efd

• [AMD][CI] Update DeepEP branch (#38396)

  Show description for 58da4ee
  rjrock

  authored
  58da4ee

  Copy full SHA for 58da4ee

• [ROCm][CI] Build fastsafetensors from source so it links against libamdhip64 (#39978)

  Show description for 1ae11e2
  AndreasKaratzas

  authored
  1ae11e2

  Copy full SHA for 1ae11e2

• skip fp8e4b15 on xpu (#39957)

  Show description for 251c18d
  

  xinyu-intel

  and

  jikunshang

  authored
  251c18d

  Copy full SHA for 251c18d

• [Misc][UX] Map mimo reasoning and tooling parsers (#40089)

  Show description for 512765d
  

  ywang96

  and

  chaunceyjiang

  authored
  512765d

  Copy full SHA for 512765d

• feat: Add LoRA support for Gemma4ForConditionalGeneration (#39291)

  Show description for 640cc9d
  

  allgather

  and

  mergify[bot]

  authored
  640cc9d

  Copy full SHA for 640cc9d