Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on May 10, 2026

• [MoE] Move various experts classes to fused_moe/experts/ (#41979)

  Show description for 1b57eb4
  

  4 peopleauthored
  1b57eb4View commit details

  Copy full SHA for 1b57eb4

  Browse repository at this point

• [Performance] Make safetensors checkpoint prefetch settings configurable (#41499)

  Show description for 21943d4
  mmangkad

  authored
  21943d4View commit details

  Copy full SHA for 21943d4

  Browse repository at this point

• [DSV4] Add PP support for deepseek-v4 (#41694)

  Show description for f396bee
  

  Isotr0py

  and

  zixi-qi

  authored
  f396beeView commit details

  Copy full SHA for f396bee

  Browse repository at this point

• [Bugfix][Mamba] IMA in causal_conv1d kernel for long sequences (#41617)

  Show description for 215e2f7
  

  Flink-ddd

  and

  mergify[bot]

  authored
  215e2f7View commit details

  Copy full SHA for 215e2f7

  Browse repository at this point

• [KV Offload] Pass ReqContext to touch(), complete_load(), and complete_store() (#41366)

  Show description for e175192
  ronensc

  authored
  e175192View commit details

  Copy full SHA for e175192

  Browse repository at this point

• [CPU] Fix spec decode kernel signatures for synthetic mode compatibility (#41932)

  Show description for a54f0d1
  

  jmamou

  and

  benchislett

  authored
  a54f0d1View commit details

  Copy full SHA for a54f0d1

  Browse repository at this point

• [Bugfix] Fuse Qwen3.5 in_qkvz_proj forwarding with LoRA enabled (#37912)

  Show description for 48698b1
  Isotr0py

  authored
  48698b1View commit details

  Copy full SHA for 48698b1

  Browse repository at this point

• [ROCm] Cap Triton paged attention block size to fix ROCm shared memory OOM (#38502)

  Show description for 0a309b5
  AndreasKaratzas

  authored
  0a309b5View commit details

  Copy full SHA for 0a309b5

  Browse repository at this point

• [CI] Trigger LoRA test when changing MoE code. (#42196)

  Show description for 84f7a55
  jeejeelee

  authored
  84f7a55View commit details

  Copy full SHA for 84f7a55

  Browse repository at this point

• [Docs] Fix broken local links (#42160)

  Show description for a2c9d54
  chfeng-cs

  authored
  a2c9d54View commit details

  Copy full SHA for a2c9d54

  Browse repository at this point

• Add @zyongye to CODEOWNERS (#42200)

  Show description for 301305c
  zyongye

  authored
  301305cView commit details

  Copy full SHA for 301305c

  Browse repository at this point

• Fix mypy failure on main (#42197)

  Show description for efd0e77
  mmangkad

  authored
  efd0e77View commit details

  Copy full SHA for efd0e77

  Browse repository at this point

• [Frontend][Bugfix] Abort ASR engine requests on cancellation (#41266)

  Show description for a5d0a5a
  

  3 peopleauthored
  a5d0a5aView commit details

  Copy full SHA for a5d0a5a

  Browse repository at this point

• [ROCm][CI] Fix NIXL spec-decode acceptance startup and diagnostics (#41313)

  Show description for f284012
  AndreasKaratzas

  authored
  f284012View commit details

  Copy full SHA for f284012

  Browse repository at this point

• [Bugfix][KV Transfer][NIXL] Notify P node on pre-admission rejection to free stranded KV blocks (#41269)

  Dao007forever

  authored
  3f5bd48View commit details

  Copy full SHA for 3f5bd48

  Browse repository at this point

• [ROCm][CI] Stabilize ROCm shutdown and distributed compile CI (#41573)

  Show description for fb1ac80
  AndreasKaratzas

  authored
  fb1ac80View commit details

  Copy full SHA for fb1ac80

  Browse repository at this point

• [Bugfix] Fix DeepSeek v4 topk numerical issue for unaligned max-model-len (#42169)

  wzhao18

  authored
  986edc8View commit details

  Copy full SHA for 986edc8

  Browse repository at this point

• docs: clarify Gemma 4 assistant speculative decoding (#42180)

  Show description for 27d3bac
  AbhiOnGithub

  authored
  27d3bacView commit details

  Copy full SHA for 27d3bac

  Browse repository at this point

• Use CU_MEMCPY_SRC_ACCESS_ORDER_ANY for batch KV cache swaps (#39306)

  Show description for 00b0618
  

  5 peopleauthored
  00b0618View commit details

  Copy full SHA for 00b0618

  Browse repository at this point

• Handle optional bool-or-string CLI args in get_kwargs (#40951)

  Show description for 0d382ec
  

  cvan20191

  and

  Christian Van

  authored
  0d382ecView commit details

  Copy full SHA for 0d382ec

  Browse repository at this point

• [CI/Build] Use modelscope's international site for regression test (#42176)

  Show description for 1029e5e
  Isotr0py

  authored
  1029e5eView commit details

  Copy full SHA for 1029e5e

  Browse repository at this point

• [Bugfix] Fix SP pass for multimodal models and PP+SP residual handling (#33322)

  Show description for 0b272a6
  

  wangxingran222

  and

  HirokenOvo

  authored
  0b272a6View commit details

  Copy full SHA for 0b272a6

  Browse repository at this point

• Fix: Nemotron 3 rescue whitespace-only final_content, not just None (#41846)

  Show description for dcb3135
  

  Naveassaf

  and

  mergify[bot]

  authored
  dcb3135View commit details

  Copy full SHA for dcb3135

  Browse repository at this point

• Add NVFP4 all-gather GEMM fusion for AsyncTP (#41882)

  Show description for bc5fdc1
  

  baonudesifeizhai

  and

  mergify[bot]

  authored
  bc5fdc1View commit details

  Copy full SHA for bc5fdc1

  Browse repository at this point

• [Bugfix] Skip routed-experts hot path when disabled (#42148)

  aoshen02

  authored
  006af4bView commit details

  Copy full SHA for 006af4b

  Browse repository at this point

Commits on May 9, 2026

• [KV Connector] Remove compat support for pre-v0.12.0 constructor signatures without `KVCacheConfig` (#39832)

  Show description for ea0e501
  yewentao256

  authored
  ea0e501View commit details

  Copy full SHA for ea0e501

  Browse repository at this point

• [Refactor] Nixl util using lazy init (#41392)

  Show description for f80aa53
  yewentao256

  authored
  f80aa53View commit details

  Copy full SHA for f80aa53

  Browse repository at this point

• [Quantization] Add ModelOpt NVFP4 W4A16 (4-bit weights, fp16/bf16 activations) support (#41769)

  Show description for 7a2b596
  juhi10071998

  authored
  7a2b596View commit details

  Copy full SHA for 7a2b596

  Browse repository at this point

• [SpecDecoding] extend mtp support for mimo 2.5 (#41905)

  Show description for 2ee8c2a
  ZJY0516

  authored
  2ee8c2aView commit details

  Copy full SHA for 2ee8c2a

  Browse repository at this point

• [Model] use AutoWeightsLoader for DeepSeekV2 (#41706)

  Show description for cd74911
  SoluMilken

  authored
  cd74911View commit details

  Copy full SHA for cd74911

  Browse repository at this point

• [BugFix] Fix Gemma4 'layers.0.moe.experts.0.down_proj_packed' KeyError issue (#40708)

  Show description for 25abddc
  SoluMilken

  authored
  25abddcView commit details

  Copy full SHA for 25abddc

  Browse repository at this point

• [Bugfix][PD] Fix DSv4 Disaggregated (#41957)

  Show description for 171d59a
  

  NickLucche

  and

  ZhanqiuHu

  authored
  171d59aView commit details

  Copy full SHA for 171d59a

  Browse repository at this point

• [Bugfix] Remove nested torch.compile in GDN rearrange_mixed_qkv causing CUDA graph capture failure (#42070)

  Show description for 3dda9ae
  

  tdoublep

  and

  ZJY0516

  authored
  3dda9aeView commit details

  Copy full SHA for 3dda9ae

  Browse repository at this point

• [Bugfix] Fix GDN KKT precision loss on Hopper GPUs by aligning tl.dot operand layout with WGMMA (#42076)

  Show description for adb6d96
  Kermit-C

  authored
  adb6d96View commit details

  Copy full SHA for adb6d96

  Browse repository at this point

• [DSv4] Improved fused Indexer Q quant kernel (#41428)

  gau-nernst

  authored
  530d371View commit details

  Copy full SHA for 530d371

  Browse repository at this point