Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit history

Commits on Jan 7, 2026

• [perf] feat: Add MFU for Qwen3-VL dense (#4753)

  Show description for 11764c6
  zhihaofang1017

  authored
  11764c6View commit details

  Copy full SHA for 11764c6

  View code at this pointBrowse repository at this point

• [ckpt] fix: FSDP save ckpt after validation (#4799)

  Show description for 8f41b05
  

  3 peopleauthored
  8f41b05View commit details

  Copy full SHA for 8f41b05

  View code at this pointBrowse repository at this point

• [doc] Update docs about fully_async_policy (#4826)

  Show description for b2205c2
  

  jsfanfanfan

  and

  jsfanfanfan

  authored
  b2205c2View commit details

  Copy full SHA for b2205c2

  View code at this pointBrowse repository at this point

• [hardware] fix: automatically set device for SFT case (#4828)

  Show description for 6484064
  A1waysBeenHere

  authored
  6484064View commit details

  Copy full SHA for 6484064

  View code at this pointBrowse repository at this point

• [doc] chore: fix checkpoint engine image link (#4821)

  Show description for 8b01d0b
  wuxibin89

  authored
  8b01d0bView commit details

  Copy full SHA for 8b01d0b

  View code at this pointBrowse repository at this point

• [trainer] feat: VeOmniEngine supports qwen3_vl ulysses (#4806)

  Show description for cf92dac
  A1waysBeenHere

  authored
  cf92dacView commit details

  Copy full SHA for cf92dac

  View code at this pointBrowse repository at this point

Commits on Jan 6, 2026

• [ckpt] feat: add checkpoint-engine abstraction (#4775)

  Show description for 6f33c55
  wuxibin89

  authored
  6f33c55View commit details

  Copy full SHA for 6f33c55

  View code at this pointBrowse repository at this point

• [vllm, sglang, rollout] fix: Fix a mistake when running run_qwen3_vl-30b-megatron.sh with latest verl and vllm0.12 (#4810)

  cboss6

  authored
  c408a6eView commit details

  Copy full SHA for c408a6e

  View code at this pointBrowse repository at this point

• [model] fix: fix temp dtype (#4813)

  Show description for 200bc6d
  

  vermouth1992

  and

  gemini-code-assist[bot]

  authored
  200bc6dView commit details

  Copy full SHA for 200bc6d

  View code at this pointBrowse repository at this point

• [recipe] feat: migrate `recipe` to the dedicated repo `verl-recipe` as a submodule (#4795)

  Show description for 2bb42ba
  tongyx361

  authored
  2bb42baView commit details

  Copy full SHA for 2bb42ba

  View code at this pointBrowse repository at this point

• [tool] fix: add tools in single_turn_agent (#4798)

  Show description for 246cac4
  Junxiao-Zhao

  authored
  246cac4View commit details

  Copy full SHA for 246cac4

  View code at this pointBrowse repository at this point

Commits on Jan 5, 2026

• [model] feat: support per sample temperature in trainer (#4787)

  Show description for f9d6198
  vermouth1992

  authored
  f9d6198View commit details

  Copy full SHA for f9d6198

  View code at this pointBrowse repository at this point

• [rollout] fix: use model_dump() for proper Pydantic serialization in token2text (#4706)

  Show description for 0f61557
  

  yurekami

  and

  claude

  authored
  0f61557View commit details

  Copy full SHA for 0f61557

  View code at this pointBrowse repository at this point

• [fsdp] feat: integrate PrefixGrouper for GRPO training acceleration (#4368)

  Show description for d6bc629
  kevssim

  authored
  d6bc629View commit details

  Copy full SHA for d6bc629

  View code at this pointBrowse repository at this point

• [ci] feat: add npu unit test (#4626)

  Show description for 3db9c99
  yyyy2000

  authored
  3db9c99View commit details

  Copy full SHA for 3db9c99

  View code at this pointBrowse repository at this point

Commits on Jan 4, 2026

• [sglang, vllm, rollout] fix: use model's max_position_embeddings for max_model_len (#4779)

  Show description for bb7fe68
  PeterSH6

  authored
  bb7fe68View commit details

  Copy full SHA for bb7fe68

  View code at this pointBrowse repository at this point

• [fsdp] feat: integrate TiledMLP for memory-efficient MLP computation (#4649)

  Show description for dcc11f2
  kevssim

  authored
  dcc11f2View commit details

  Copy full SHA for dcc11f2

  View code at this pointBrowse repository at this point

• [worker] fix: Model engine parameter offload (#4777)

  Show description for 681af5f
  JacobHelwig

  authored
  681af5fView commit details

  Copy full SHA for 681af5f

  View code at this pointBrowse repository at this point

• [cfg] chore: remove redundant fields and fix typo (#4754)

  Show description for c191c5e
  JoyboyBrian

  authored
  c191c5eView commit details

  Copy full SHA for c191c5e

  View code at this pointBrowse repository at this point

Commits on Jan 3, 2026

• [worker] fix: Config for PPO batch size (#4773)

  Show description for 4e40245
  JacobHelwig

  authored
  4e40245View commit details

  Copy full SHA for 4e40245

  View code at this pointBrowse repository at this point

• [training_utils] fix: Nested tensor micro-batching (#4776)

  Show description for 175d5b8
  JacobHelwig

  authored
  175d5b8View commit details

  Copy full SHA for 175d5b8

  View code at this pointBrowse repository at this point

• [trainer] feat: enable ray-based sft trainer on ascend npu (#4764)

  Show description for 9550a6b
  ji-huazhong

  authored
  9550a6bView commit details

  Copy full SHA for 9550a6b

  View code at this pointBrowse repository at this point

• [perf] feat: add remote reward manager and fix math verify issue (#4752)

  Show description for 7c42510
  yyDing1

  authored
  7c42510View commit details

  Copy full SHA for 7c42510

  View code at this pointBrowse repository at this point

• [data] feat: major refactor RLHFDataset for multi-modal data (#4759)

  Show description for 51d2737
  wuxibin89

  authored
  51d2737View commit details

  Copy full SHA for 51d2737

  View code at this pointBrowse repository at this point

Commits on Jan 2, 2026

• Revert "[algo] fix: Add seq mean mask denominator option" (#4769)

  Show description for 78014a2
  wuxibin89

  authored
  78014a2View commit details

  Copy full SHA for 78014a2

  View code at this pointBrowse repository at this point

• [vllm] fix: use packaging.version for correct semantic version comparison (#4768)

  Racktic

  authored
  380d373View commit details

  Copy full SHA for 380d373

  View code at this pointBrowse repository at this point

• [doc, cfg] fix: correct typos in training and docker configurations (#4767)

  Racktic

  authored
  4827f76View commit details

  Copy full SHA for 4827f76

  View code at this pointBrowse repository at this point

Commits on Jan 1, 2026

• [ci] fix: fix precommit (#4760)

  vermouth1992

  authored
  b4d1050View commit details

  Copy full SHA for b4d1050

  View code at this pointBrowse repository at this point

Commits on Dec 31, 2025

• [rollout] feat: Add vllm logprob mode and default processed_logprob (#4755)

  RobotGF

  authored
  7a82f2eView commit details

  Copy full SHA for 7a82f2e

  View code at this pointBrowse repository at this point

• [trainer] fix: use dp_size instead of world_size in _balance_batch (#4697)

  yurekami

  authored
  b16e048View commit details

  Copy full SHA for b16e048

  View code at this pointBrowse repository at this point

• [tool] fix: attach to existing MLflow run when MLFLOW_RUN_ID is set (#4740)

  dubin555

  authored
  2d02788View commit details

  Copy full SHA for 2d02788

  View code at this pointBrowse repository at this point

• [perf]feat: GPT-OSS mfu compute support (#4750)

  mikequan0425

  authored
  854ca34View commit details

  Copy full SHA for 854ca34

  View code at this pointBrowse repository at this point

• [trainer,megatron] fix: super tiny fix the issue of repeatedly importing the mindspeed patch (#4751)

  ji-huazhong

  authored
  4017d63View commit details

  Copy full SHA for 4017d63

  View code at this pointBrowse repository at this point

• [perf] feat: support profiler in model engine and sft trainer (#4749)

  Show description for 0a5e856
  vermouth1992

  authored
  0a5e856View commit details

  Copy full SHA for 0a5e856

  View code at this pointBrowse repository at this point

• [reward] fix: make `RateLimitedRewardManager` accept legacy kwargs (#4739)

  Show description for a090cd8
  JoyboyBrian

  authored
  a090cd8View commit details

  Copy full SHA for a090cd8

  View code at this pointBrowse repository at this point