Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit history

Commits on Feb 20, 2026

• [trainer] feat: Add Torchtitan as alternative training engine (#5051)

  Show description for f5c34bb
  acisseJZhong

  authored
  f5c34bbView commit details

  Copy full SHA for f5c34bb

  View code at this pointBrowse repository at this point

Commits on Feb 14, 2026

• [fsdp,algo] feat: add NVFP4 QAT (Quantization-Aware Training) support (#5190)

  Show description for 27f65f1
  zhangyimi

  authored
  27f65f1View commit details

  Copy full SHA for 27f65f1

  View code at this pointBrowse repository at this point

Commits on Feb 13, 2026

• [veomni] refactor: Modify dp related parameters to align with FSDP backend and remove temporarily unsupported TP/PP/CP parameters (#5303)

  Show description for a43eecb
  

  ChengQianqian

  and

  ji-huazhong

  authored
  a43eecbView commit details

  Copy full SHA for a43eecb

  View code at this pointBrowse repository at this point

Commits on Feb 12, 2026

• [reward] fix: reward model args and reward_kwargs bug (#5289)

  Show description for 92c454a
  yyDing1

  authored
  92c454aView commit details

  Copy full SHA for 92c454a

  View code at this pointBrowse repository at this point

Commits on Feb 11, 2026

• [reward] fix: backward compatibility with old reward config (#5287)

  Show description for 018f8dc
  yyDing1

  authored
  018f8dcView commit details

  Copy full SHA for 018f8dc

  View code at this pointBrowse repository at this point

• [misc] chore: fix veomni_trainer.yaml (#5285)

  Show description for 12b6382
  wuxibin89

  authored
  12b6382View commit details

  Copy full SHA for 12b6382

  View code at this pointBrowse repository at this point

• [BREAKING][reward] refactor: the full reward configuration (#5255)

  Show description for 67dc84b
  yyDing1

  authored
  67dc84bView commit details

  Copy full SHA for 67dc84b

  View code at this pointBrowse repository at this point

Commits on Feb 10, 2026

• [veomni, trainer] feat: add rl support for veomni backend (#4882)

  Show description for 2487c36
  

  ji-huazhong

  and

  A1waysBeenHere

  authored
  2487c36View commit details

  Copy full SHA for 2487c36

  View code at this pointBrowse repository at this point

Commits on Feb 9, 2026

• [BREAKING][reward] refactor: deprecate batch reward manager (#5237)

  Show description for 3df21a6
  yyDing1

  authored
  3df21a6View commit details

  Copy full SHA for 3df21a6

  View code at this pointBrowse repository at this point

• [megatron] feat: use yaml to manage mbridge args (#4584)

  Show description for 0b0769e
  

  3 peopleauthored
  0b0769eView commit details

  Copy full SHA for 0b0769e

  View code at this pointBrowse repository at this point

Commits on Feb 6, 2026

• [BREAKING][reward] refactor: remove reward model worker code and invocation (#5194)

  Show description for 4ef45d0
  yyDing1

  authored
  4ef45d0View commit details

  Copy full SHA for 4ef45d0

  View code at this pointBrowse repository at this point

• [algo] feat: support rollout router replay in MegatronEngine (#5185)

  Show description for d987199
  xhx1022

  authored
  d987199View commit details

  Copy full SHA for d987199

  View code at this pointBrowse repository at this point

Commits on Feb 4, 2026

• [perf] fix: infer server profiler args fix (#5121)

  Show description for e4915bd
  mengchengTang

  authored
  e4915bdView commit details

  Copy full SHA for e4915bd

  View code at this pointBrowse repository at this point

Commits on Feb 3, 2026

• [megatron] fix: checkpoints uses `fully_reshardable` by default when supported (#5154)

  Show description for ded57b0
  HollowMan6

  authored
  ded57b0View commit details

  Copy full SHA for ded57b0

  View code at this pointBrowse repository at this point

Commits on Feb 2, 2026

• [reward] feat: split reward loop manager and agent loop manager (#5134)

  Show description for 84d9d8f
  yyDing1

  authored
  84d9d8fView commit details

  Copy full SHA for 84d9d8f

  View code at this pointBrowse repository at this point

Commits on Jan 31, 2026

• [rollout,vllm] feat: revert default value of max_num_seqs (#5139)

  Show description for e2ae919
  RobotGF

  authored
  e2ae919View commit details

  Copy full SHA for e2ae919

  View code at this pointBrowse repository at this point

Commits on Jan 30, 2026

• [sglang] fix: update wiki to support speculative decode rollout (#5116)

  Show description for 8779775
  ArronHZG

  authored
  8779775View commit details

  Copy full SHA for 8779775

  View code at this pointBrowse repository at this point

Commits on Jan 29, 2026

• [megatron] feat: Support MTP training in SFT (#4981)

  Show description for adff795
  

  3 peopleauthored
  adff795View commit details

  Copy full SHA for adff795

  View code at this pointBrowse repository at this point

Commits on Jan 27, 2026

• [ckpt] feat: add CheckpointEngineManager (#5031)

  Show description for ef6eaa0
  wuxibin89

  authored
  ef6eaa0View commit details

  Copy full SHA for ef6eaa0

  View code at this pointBrowse repository at this point

Commits on Jan 24, 2026

• [veomni] refactor: no long check the attn_implementation/moe_implementation in VeOmniEngineConfig (#5019)

  Show description for 772c224
  

  A1waysBeenHere

  and

  ji-huazhong

  authored
  772c224View commit details

  Copy full SHA for 772c224

  View code at this pointBrowse repository at this point

• [megatron] feat: LoRA adapter only refit (TensorLoRARequest) (#4632)

  Show description for b703663
  HollowMan6

  authored
  b703663View commit details

  Copy full SHA for b703663

  View code at this pointBrowse repository at this point

Commits on Jan 23, 2026

• [BREAKING][worker, rollout, vllm] feat: implement vLLM colocated training-inference rollout with process separation (#4280)

  Show description for e9405d7
  

  jianjunzhong

  and

  wuxibin89

  authored
  e9405d7View commit details

  Copy full SHA for e9405d7

  View code at this pointBrowse repository at this point

Commits on Jan 21, 2026

• [perf] feat: verl profiler system support Agent Loop scenario and integrate torch.profiler (#4320)

  Show description for cd1926c
  

  mengchengTang

  and

  tardis-key

  authored
  cd1926cView commit details

  Copy full SHA for cd1926c

  View code at this pointBrowse repository at this point

Commits on Jan 20, 2026

• [ray,rollout,trtllm] feat: Adding tensorrt_llm as new rollout engine (#4665)

  Show description for f65fd72
  

  joyang-nv

  and

  Superjomn

  authored
  f65fd72View commit details

  Copy full SHA for f65fd72

  View code at this pointBrowse repository at this point

Commits on Jan 19, 2026

• [megatron] feat: Using MTP in RL Training and Inference (#4936)

  Show description for df340d7
  ArronHZG

  authored
  df340d7View commit details

  Copy full SHA for df340d7

  View code at this pointBrowse repository at this point

Commits on Jan 16, 2026

• [veomni] feat: support offloading/loading the veomni model/optimizer (#4916)

  Show description for f98fed1
  ji-huazhong

  authored
  f98fed1View commit details

  Copy full SHA for f98fed1

  View code at this pointBrowse repository at this point

Commits on Jan 14, 2026

• [algo, doc] feat: trust region sequence masking - (1) k3 KL avg and (2) veto for max criterion (#4544)

  Show description for 97eac90
  

  3 peopleauthored
  97eac90View commit details

  Copy full SHA for 97eac90

  View code at this pointBrowse repository at this point

Commits on Jan 11, 2026

• [rollout,vllm] feat: support vllm scheduling policy config and generate setting priority (#4874)

  Show description for bc73797
  

  RobotGF

  and

  juliusgao

  authored
  bc73797View commit details

  Copy full SHA for bc73797

  View code at this pointBrowse repository at this point

Commits on Jan 7, 2026

• [algo] feat: add optimal token baseline and variance proxy (#4678)

  Show description for 67a127b
  

  3 peopleauthored
  67a127bView commit details

  Copy full SHA for 67a127b

  View code at this pointBrowse repository at this point

Commits on Jan 5, 2026

• [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

Commits on Jan 4, 2026

• [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

Commits on Jan 2, 2026

• [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 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

• [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

Commits on Dec 30, 2025

• [rollout,docs] fix: improve error message (#4682) and docstrings (#1345) (#4729)

  Show description for f53e2e5
  

  yurekami

  and

  claude

  authored
  f53e2e5View commit details

  Copy full SHA for f53e2e5

  View code at this pointBrowse repository at this point