Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit history

Commits on Sep 29, 2025

• [sglang] feat: add preparation for sglang+verl (#3506)

  Show description for f50e5c2
  

  lbk-sys

  and

  1StepForever

  authored
  f50e5c2View commit details

  Copy full SHA for f50e5c2

  View code at this pointBrowse repository at this point

Commits on Sep 28, 2025

• [megatron, training_utils] fix: encoder pp is removed in mcore >= 0.14 (#3640)

  Show description for 9e2072d
  HollowMan6

  authored
  9e2072dView commit details

  Copy full SHA for 9e2072d

  View code at this pointBrowse repository at this point

• [rollout,vllm] fix: Add LoRA Loading to Async vLLM (#3639)

  Show description for 39e531f
  kfallah

  authored
  39e531fView commit details

  Copy full SHA for 39e531f

  View code at this pointBrowse repository at this point

• [megatron, worker] fix: use `extract_multi_modal_inputs` method for handling `multi_modal_inputs` (#3641)

  Show description for abca659
  HollowMan6

  authored
  abca659View commit details

  Copy full SHA for abca659

  View code at this pointBrowse repository at this point

Commits on Sep 27, 2025

• [algo, perf] feat: Vectorize GRPO Advantage Estimator - 13～26x Speedup (#3635)

  Show description for 4ff3ce2
  CedricHwong

  authored
  4ff3ce2View commit details

  Copy full SHA for 4ff3ce2

  View code at this pointBrowse repository at this point

Commits on Sep 26, 2025

• [model] feat: add glm4v (#3291)

  Show description for c03dcb0
  

  5 peopleauthored
  c03dcb0View commit details

  Copy full SHA for c03dcb0

  View code at this pointBrowse repository at this point

• [2/N][rollout] feat: support vllm/sglang DP+EP in server mode (#3530)

  Show description for 84d5619
  wuxibin89

  authored
  84d5619View commit details

  Copy full SHA for 84d5619

  View code at this pointBrowse repository at this point

• [trainer] fix: Ref to #3596. More import fix for transformers version higher than 4.55.0 (#3608)

  Show description for 64a9860
  A1waysBeenHere

  authored
  64a9860View commit details

  Copy full SHA for 64a9860

  View code at this pointBrowse repository at this point

• [rollout] refactor: Update rollout and reward configs to reuse vllm/sglang replicas (#3625)

  Show description for e513058
  yyDing1

  authored
  e513058View commit details

  Copy full SHA for e513058

  View code at this pointBrowse repository at this point

• [megatron] feat: add mindspeed engine and support sft (#3599)

  Show description for 2234810
  

  ji-huazhong

  and

  baymax591

  authored
  2234810View commit details

  Copy full SHA for 2234810

  View code at this pointBrowse repository at this point

• [recipe] fix: Fix a Typo in One_Step_Off_Policy and Add async of Generative Reward Model in Response Generation (#3369)

  Show description for 377bbb8
  

  ZhichaoWang970201

  and

  Zhichao-Jimmy970201

  authored
  377bbb8View commit details

  Copy full SHA for 377bbb8

  View code at this pointBrowse repository at this point

• [ci] feat: upgrade sglang to 0.5.2 (#3613)

  Show description for 6ff2b43
  wuxibin89

  authored
  6ff2b43View commit details

  Copy full SHA for 6ff2b43

  View code at this pointBrowse repository at this point

Commits on Sep 25, 2025

• [rollout, tool] feat: export rollout rewards to total rewards (#3563)

  Show description for bf7aac2
  Tavish9

  authored
  bf7aac2View commit details

  Copy full SHA for bf7aac2

  View code at this pointBrowse repository at this point

• [worker] fix: correctly determine is_vlm_model if sp > 1 (#3282)

  Show description for 616e933
  HollowMan6

  authored
  616e933View commit details

  Copy full SHA for 616e933

  View code at this pointBrowse repository at this point

• [rollout] fix: remove code responsible for tool response duplication (#3604)

  Show description for 7731c5c
  mgilmore-relace

  authored
  7731c5cView commit details

  Copy full SHA for 7731c5c

  View code at this pointBrowse repository at this point

• [recipe] feat: CollabLLM integration for multiturn training (#3574)

  Show description for 25d78fa
  

  Wuyxin

  and

  chenhaiq

  authored
  25d78faView commit details

  Copy full SHA for 25d78fa

  View code at this pointBrowse repository at this point

Commits on Sep 24, 2025

• [trainer] fix: Import flash attn utils for Transformers higher than 4.55.0 (#3596)

  Show description for ba85551
  

  3 peopleauthored
  ba85551View commit details

  Copy full SHA for ba85551

  View code at this pointBrowse repository at this point

• [algo, perf] feat: Vectorize RLOO Advantage Estimator - 20x Speedup (#3555)

  Show description for 26a734e
  EduardDurech

  authored
  26a734eView commit details

  Copy full SHA for 26a734e

  View code at this pointBrowse repository at this point

• [misc] feat: prototype deprecate DataProto and replace with Tensordict: part 2 (#3567)

  Show description for 69b0127
  

  houminz

  and

  vermouth1992

  authored
  69b0127View commit details

  Copy full SHA for 69b0127

  View code at this pointBrowse repository at this point

• [megatron] fix: revert megatron actor refactor (#3553)

  Show description for 1985eb1
  vermouth1992

  authored
  1985eb1View commit details

  Copy full SHA for 1985eb1

  View code at this pointBrowse repository at this point

• [megatron] feat: use flash as default attention_backend (#3578)

  Show description for f1d212c
  ISEEKYAN

  authored
  f1d212cView commit details

  Copy full SHA for f1d212c

  View code at this pointBrowse repository at this point

Commits on Sep 23, 2025

• [sglang] fix: Support SGLang>=0.5.2 (#3526)

  Show description for aaa4cf5
  EduardDurech

  authored
  aaa4cf5View commit details

  Copy full SHA for aaa4cf5

  View code at this pointBrowse repository at this point

• [megatron] fix: fix bug when holding empty parameters with custom pipeline layout (#3565)

  Show description for 0603b7b
  

  HaochenYuan

  and

  gemini-code-assist[bot]

  authored
  0603b7bView commit details

  Copy full SHA for 0603b7b

  View code at this pointBrowse repository at this point

• [misc] feat: remove redundant default params (#3577)

  Show description for 5150686
  

  4 peopleauthored
  5150686View commit details

  Copy full SHA for 5150686

  View code at this pointBrowse repository at this point

• [CI] chore: Update e2e_ascend CI config (#3532)

  Show description for 5547dbe
  FightingZhen

  authored
  5547dbeView commit details

  Copy full SHA for 5547dbe

  View code at this pointBrowse repository at this point

• [trainer] refactor: move rollout log to inheritable trainer (#3576)

  Show description for 50368ae
  ccclyu

  authored
  50368aeView commit details

  Copy full SHA for 50368ae

  View code at this pointBrowse repository at this point

Commits on Sep 22, 2025

• [trainer,rollout] fix: ensure LoRA weights are loaded when vllm_sleep_level=2 and without using layerd_summon (#3541)

  Show description for 96e7071
  piood

  authored
  96e7071View commit details

  Copy full SHA for 96e7071

  View code at this pointBrowse repository at this point

Commits on Sep 19, 2025

• [model] feat: support parameter generator for model engine (#3529)

  vermouth1992

  authored
  d33c85eView commit details

  Copy full SHA for d33c85e

  View code at this pointBrowse repository at this point

• [Megatron] fix: compatible to mcore0.15 (#3534)

  Show description for bbdf819
  ISEEKYAN

  authored
  bbdf819View commit details

  Copy full SHA for bbdf819

  View code at this pointBrowse repository at this point

• [worker] fix: get all `multi_modal_inputs` keys with in a microbatch (#3315)

  Show description for 2f6a5d6
  HollowMan6

  authored
  2f6a5d6View commit details

  Copy full SHA for 2f6a5d6

  View code at this pointBrowse repository at this point

Commits on Sep 18, 2025

• [model] fix: qwen2vl for transformers 4.52.* (#3524)

  hiyouga

  authored
  c0e2b9dView commit details

  Copy full SHA for c0e2b9d

  View code at this pointBrowse repository at this point

• [megatron] Add TIS support to megatron backend (#3513)

  Show description for b6b34b2
  

  sharonyu-115

  and

  Shuang Yu

  authored
  b6b34b2View commit details

  Copy full SHA for b6b34b2

  View code at this pointBrowse repository at this point

• [model] fix: refactor qwen2vl patches & support no-image input for fsdp (#3496)

  Show description for 0d4541f
  hiyouga

  authored
  0d4541fView commit details

  Copy full SHA for 0d4541f

  View code at this pointBrowse repository at this point

Commits on Sep 17, 2025

• [data] feat: support customizable loss mask in multi-turn sft dataset (#3507)

  Show description for 214d0f0
  vermouth1992

  authored
  214d0f0View commit details

  Copy full SHA for 214d0f0

  View code at this pointBrowse repository at this point

• [training_utils] refactor: extract checkpoint handler into a separate file for reuse (#3505)

  Show description for 0665153
  vermouth1992

  authored
  0665153View commit details

  Copy full SHA for 0665153

  View code at this pointBrowse repository at this point