Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit history

Commits on Jul 10, 2026

• feat: add dataset-specific batch LLM judge for online DPO

  Show description for 1e012cc
  

  lvhaoyu

  and

  claude

  committed
  1e012ccView commit details

  Copy full SHA for 1e012cc

  View code at this pointBrowse repository at this point

• support uniagent for dpo

  

  lvhaoyu

  committed
  f7cdc12View commit details

  Copy full SHA for f7cdc12

  View code at this pointBrowse repository at this point

Commits on Jul 8, 2026

• update batch dpo with model judge

  

  lvhaoyu

  committed
  d12964dView commit details

  Copy full SHA for d12964d

  View code at this pointBrowse repository at this point

• update

  

  lvhaoyu

  committed
  e36e338View commit details

  Copy full SHA for e36e338

  View code at this pointBrowse repository at this point

• add agent online dpo

  

  lvhaoyu

  committed
  f47a653View commit details

  Copy full SHA for f47a653

  View code at this pointBrowse repository at this point

• update

  

  lvhaoyu

  committed
  b1f5cc9View commit details

  Copy full SHA for b1f5cc9

  View code at this pointBrowse repository at this point

• add optimizer reset per iter feature

  

  lvhaoyu

  committed
  d57fceaView commit details

  Copy full SHA for d57fcea

  View code at this pointBrowse repository at this point

• Refactor math_reward.py for clarity and efficiency

  Show description for edbd3d5
  

  frelam

  authored and

  lvhaoyu

  committed
  edbd3d5View commit details

  Copy full SHA for edbd3d5

  View code at this pointBrowse repository at this point

• Refactor string processing in math_reward.py

  

  frelam

  authored and

  lvhaoyu

  committed
  b3f47dcView commit details

  Copy full SHA for b3f47dc

  View code at this pointBrowse repository at this point

• fix muon

  

  lvhaoyu

  committed
  bede00eView commit details

  Copy full SHA for bede00e

  View code at this pointBrowse repository at this point

• fix muon

  

  lvhaoyu

  committed
  84289d4View commit details

  Copy full SHA for 84289d4

  View code at this pointBrowse repository at this point

• fix muon optimizer

  

  lvhaoyu

  committed
  f7c4aa2View commit details

  Copy full SHA for f7c4aa2

  View code at this pointBrowse repository at this point

• add muon optimizer

  

  lvhaoyu

  committed
  9c4b471View commit details

  Copy full SHA for 9c4b471

  View code at this pointBrowse repository at this point

• add length penalty for gdpo

  

  lvhaoyu

  committed
  8161c7fView commit details

  Copy full SHA for 8161c7f

  View code at this pointBrowse repository at this point

• add deepseek v3.2 offpolicy_seq_mask

  

  lvhaoyu

  committed
  7cef4bdView commit details

  Copy full SHA for 7cef4bd

  View code at this pointBrowse repository at this point

• add reward and train receipe for math specilist

  

  lvhaoyu

  committed
  e9a7681View commit details

  Copy full SHA for e9a7681

  View code at this pointBrowse repository at this point

• add deepseek v3.2 unbias KL Estimation and keep topp topk masking

  

  lvhaoyu

  committed
  119d78cView commit details

  Copy full SHA for 119d78c

  View code at this pointBrowse repository at this point

• update math reward

  

  lvhaoyu

  committed
  5eaf188View commit details

  Copy full SHA for 5eaf188

  View code at this pointBrowse repository at this point

• update gdpo score

  

  lvhaoyu

  committed
  f33b788View commit details

  Copy full SHA for f33b788

  View code at this pointBrowse repository at this point

• add gdpo custom reward

  

  lvhaoyu

  committed
  13c93f0View commit details

  Copy full SHA for 13c93f0

  View code at this pointBrowse repository at this point

• update seq mean for kl penalty in reward

  

  lvhaoyu

  committed
  5acbcdcView commit details

  Copy full SHA for 5acbcdc

  View code at this pointBrowse repository at this point

• add repetition_penalty for rollout

  

  lvhaoyu

  committed
  1619aadView commit details

  Copy full SHA for 1619aad

  View code at this pointBrowse repository at this point

• add length penalty for dapo only when score > 0

  

  lvhaoyu

  committed
  e97342eView commit details

  Copy full SHA for e97342e

  View code at this pointBrowse repository at this point

• fix mathreward

  

  lvhaoyu

  committed
  f8e91bdView commit details

  Copy full SHA for f8e91bd

  View code at this pointBrowse repository at this point

• add dynamic sample algorithum

  

  lvhaoyu

  committed
  1783761View commit details

  Copy full SHA for 1783761

  View code at this pointBrowse repository at this point

• add custom mix math reward

  

  lvhaoyu

  committed
  51728d1View commit details

  Copy full SHA for 51728d1

  View code at this pointBrowse repository at this point

• add kl masked wrong sample and dynamic sample hard sample

  

  lvhaoyu

  committed
  3ab49f3View commit details

  Copy full SHA for 3ab49f3

  View code at this pointBrowse repository at this point

• add mix dataset

  

  lvhaoyu

  committed
  ebbfe15View commit details

  Copy full SHA for ebbfe15

  View code at this pointBrowse repository at this point

Commits on Jul 7, 2026

• [ckpt] fix: use separate magic_recv buffer to prevent weight corruption (#6813)

  Show description for 7b19203
  KunWuLuan

  authored
  7b19203View commit details

  Copy full SHA for 7b19203

  View code at this pointBrowse repository at this point

• [perf] feat: Add Qwen3.5 MFU flops estimation (#6389)

  Show description for dbd307e
  liangxuZhang

  authored
  dbd307eView commit details

  Copy full SHA for dbd307e

  View code at this pointBrowse repository at this point

• [algo] fix: normalize critic value loss over the global mini-batch, not per micro-batch (#6957)

  Show description for 2eb020a
  abtonmoy

  authored
  2eb020aView commit details

  Copy full SHA for 2eb020a

  View code at this pointBrowse repository at this point

• [megatron] fix: free Transformer-Engine FP8 weight workspaces on CPU offload (#6952)

  Show description for e52747a
  

  4 peopleauthored
  e52747aView commit details

  Copy full SHA for e52747a

  View code at this pointBrowse repository at this point

• [fsdp] fix: fused-kernel gradient tracking bug. (#6913)

  Show description for e48725a
  

  kolehma8

  and

  gemini-code-assist[bot]

  authored
  e48725aView commit details

  Copy full SHA for e48725a

  View code at this pointBrowse repository at this point

• [vllm, rollout] fix: require vLLM>=0.22.0 for Qwen3.5 router replay (R3) (#6922)

  Show description for 2ef858e
  dafu-wu

  authored
  2ef858eView commit details

  Copy full SHA for 2ef858e

  View code at this pointBrowse repository at this point

• [trainer] fix: spmd_types and activation checkpointing composability bug (#6950)

  Show description for ef500de
  acisseJZhong

  authored
  ef500deView commit details

  Copy full SHA for ef500de

  View code at this pointBrowse repository at this point