Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Jun 12, 2026

• [reward] fix: add optional per-sample compute_score timeout to NaiveRewardManager (#6673)

  Show description for 9f18abe
  zqinglin

  authored
  9f18abe

  Copy full SHA for 9f18abe

• [hardware] feat: add ROCm/HIP platform backend (PlatformROCm) (#6702)

  Show description for ab8da13
  

  xiaohong42

  and

  cursoragent

  authored
  ab8da13

  Copy full SHA for ab8da13

• [rollout] fix: clone LoRA weights out of the reused IPC buffer before add_lora (#6688)

  Show description for 173945f
  

  HaozheZhang6

  and

  claude

  authored
  173945f

  Copy full SHA for 173945f

• [fsdp] fix: handle missing chunked top-k config (#6704)

  Show description for 4a0ff06
  Luosuu

  authored
  4a0ff06

  Copy full SHA for 4a0ff06

• [fsdp, trainer] fix: detach model_output and loss metrics to stop per-micro-batch graph retention in actor update (#6699)

  Show description for dbbf085
  YoungZSh

  authored
  dbbf085

  Copy full SHA for dbbf085

• [ci] test: add qwen3.5 2b ascend nightly ci (#6637)

  Show description for bb3d9b4
  Zhang1Sheng

  authored
  bb3d9b4

  Copy full SHA for bb3d9b4

• [ci] chore: add npu's veomni ci (#6701)

  Show description for 5e07096
  daikang6

  authored
  5e07096

  Copy full SHA for 5e07096

Commits on Jun 10, 2026

• [megatron] fix: propagate disabled MTP config to bridge providers (#6626)

  Show description for 41a5244
  Mengyuyang

  authored
  41a5244

  Copy full SHA for 41a5244

• [rollout] fix: propagate ignore_eos to vLLM sampling params (#6630)

  Show description for 15560ea
  Mengyuyang

  authored
  15560ea

  Copy full SHA for 15560ea

• [trainer, megatron] fix: fix MindSpeed repatch config bug (#6674)

  Show description for a1fad97
  chengminhua

  authored
  a1fad97

  Copy full SHA for a1fad97

• [ci, hardware] feat: add AMD ROCm (MI300) e2e_ppo_trainer workflow (#6668)

  Show description for d82d277
  

  3 peopleauthored
  d82d277

  Copy full SHA for d82d277

• [ci] fix: ignore one ut which failed in npu ci (#6679)

  Show description for e5d1784
  yyyy2000

  authored
  e5d1784

  Copy full SHA for e5d1784

• [fsdp] feat: chunked gather-logsumexp for top-K loss to avoid OOM at long context (#6593)

  kekellllll

  authored
  c6144ec

  Copy full SHA for c6144ec

• [veomni, cfg] feat: add missing config fields to veomni.yaml (#6505)

  Show description for 5f74f82
  mikequan0425

  authored
  5f74f82

  Copy full SHA for 5f74f82

Commits on Jun 9, 2026

• [doc] feat: update Qwen3.5 NPU guide (#6667)

  Show description for 4edf977
  zjchenn

  authored
  4edf977

  Copy full SHA for 4edf977

• [doc] chore: Update ascend_vllm_best_practices.rst (#6640)

  Show description for 5a38699
  

  3 peopleauthored
  5a38699

  Copy full SHA for 5a38699

• [vllm, fsdp] fix: apply FSDP buffer updates during rollout weight sync (#5801)

  Show description for a1f6e71
  chenshui223

  authored
  a1f6e71

  Copy full SHA for a1f6e71

Commits on Jun 8, 2026

• [veomni] fix: add veomni backend docs for ascend (#6657)

  Show description for 29ffe75
  phdddd

  authored
  29ffe75

  Copy full SHA for 29ffe75

• [vllm] fix: use data-parallel rank in vLLM ZMQ handles (#6620)

  Show description for 478b718
  he-yufeng

  authored
  478b718

  Copy full SHA for 478b718

• [megatron] fix: MTP compatible with latest mcore (#6648)

  Show description for f65b2ce
  HollowMan6

  authored
  f65b2ce

  Copy full SHA for f65b2ce

• [data] chore: Update TransferQueue version to 0.1.8 (#6652)

  Show description for cea8661
  0oshowero0

  authored
  cea8661

  Copy full SHA for cea8661

• [data] fix: support path multimodal placeholders (#6631)

  Show description for f48b143
  le-czs

  authored
  f48b143

  Copy full SHA for f48b143

• [megatron] fix: Fix Megatron optimizer offloading when `use_precision_aware_optimizer=True` (#6635)

  Show description for 587b120
  

  rao-ashish

  and

  gemini-code-assist[bot]

  authored
  587b120

  Copy full SHA for 587b120

• [megatron] fix: official moe_enable_routing_replay flag in addition (#6647)

  Show description for 62197aa
  HollowMan6

  authored
  62197aa

  Copy full SHA for 62197aa

• [ci] fix: several CI syntax errors (#6649)

  Show description for a568bfd
  HollowMan6

  authored
  a568bfd

  Copy full SHA for a568bfd

• [veomni] fix: pass mixed precision config to veomni (#6644)

  Show description for 9add650
  

  Luosuu

  and

  codex

  authored
  9add650

  Copy full SHA for 9add650

Commits on Jun 7, 2026

• [rollout] fix: avoid arbitrary code execution in Qwen3 tool parser (#6542)

  Show description for 2084d8f
  

  fuergaosi233

  and

  claude

  authored
  2084d8f

  Copy full SHA for 2084d8f

Commits on Jun 6, 2026

• [ci] fix: Modify the log path of npu nightly ci (#6600)

  Show description for 4ff9ca5
  daikang6

  authored
  4ff9ca5

  Copy full SHA for 4ff9ca5

• [fsdp] feat: add Qwen3.5-4B on-policy distillation FSDP script (#6638)

  Show description for 7757fb0
  cls1206

  authored
  7757fb0

  Copy full SHA for 7757fb0

• [ci] fix: remove uv usage from A2 vlm_rl_job, use pip instead (#6618)

  Show description for 68930b6
  

  3 peopleauthored
  68930b6

  Copy full SHA for 68930b6

Commits on Jun 5, 2026

• [megatron] fix: clamp num_tokens=0 in MTP loss & add normalized scale for MTP per token loss (#6464)

  Show description for 8911f24
  arvyanh

  authored
  8911f24

  Copy full SHA for 8911f24

• [misc] fix: revert run_qwen3_8b_fsdp_npu.sh (#6601)

  Show description for c131c70
  hustmf

  authored
  c131c70

  Copy full SHA for c131c70

• [doc] chore: update amd rocm index (#6622)

  Show description for 189663a
  wuxibin89

  authored
  189663a

  Copy full SHA for 189663a

• [doc] chore: Update index.rst (#6617)

  Show description for 6209469
  hustmf

  authored
  6209469

  Copy full SHA for 6209469

• [fsdp] fix: do not manually move model to GPU (#6604)

  Show description for a539474
  

  pavelgein

  and

  gemini-code-assist[bot]

  authored
  a539474

  Copy full SHA for a539474