Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit history

Commits on Mar 17, 2025

• [rollout] feat: add SGLang as rollout engine to verl (#490)

  Show description for 333e6d6
  

  ocss884

  and

  zhaochenyang20

  authored
  333e6d6View commit details

  Copy full SHA for 333e6d6

  View code at this pointBrowse repository at this point

Commits on Mar 15, 2025

• [misc] fix: validation batch repeat before feed into rollout (#614)

  PeterSH6

  authored
  cb943beView commit details

  Copy full SHA for cb943be

  View code at this pointBrowse repository at this point

• misc: separate metric utils from ppo trainer (#599)

  Show description for 6133ae9
  hiyouga

  authored
  6133ae9View commit details

  Copy full SHA for 6133ae9

  View code at this pointBrowse repository at this point

Commits on Mar 14, 2025

• Support for GRPO with Megatron backend (#592)

  Show description for c342069
  ETOgaosion

  authored
  c342069View commit details

  Copy full SHA for c342069

  View code at this pointBrowse repository at this point

• fix: Add error mechanism for mini-batch/batch size divisibility validation (#559)

  fyqqyf

  authored
  4d27461View commit details

  Copy full SHA for 4d27461

  View code at this pointBrowse repository at this point

• [config] feat: lr_warmup_steps (#564)

  Show description for 22657ba
  tongyx361

  authored
  22657baView commit details

  Copy full SHA for 22657ba

  View code at this pointBrowse repository at this point

• [update] delete useless config params (#591)

  BearBiscuit05

  authored
  99ea19aView commit details

  Copy full SHA for 99ea19a

  View code at this pointBrowse repository at this point

• [Config] Providing an option to turn off `torch.compile` in actor (#554)

  Show description for 5457469
  hongpeng-guo

  authored
  5457469View commit details

  Copy full SHA for 5457469

  View code at this pointBrowse repository at this point

• misc: remove redundant .to(device) (#565)

  Show description for d2db725
  dirtyDan0

  authored
  d2db725View commit details

  Copy full SHA for d2db725

  View code at this pointBrowse repository at this point

Commits on Mar 13, 2025

• fix bug #544 that 'left' and 'right' config for truncation don't work (#583)

  none0663

  authored
  e7c40b3View commit details

  Copy full SHA for e7c40b3

  View code at this pointBrowse repository at this point

• fix: remove redundant broadcast in fsdp vllm postprocess (#577)

  Show description for f7e183e
  wuxibin89

  authored
  f7e183eView commit details

  Copy full SHA for f7e183e

  View code at this pointBrowse repository at this point

• fix: remove redundant torch.cuda.empty_cache() (#575)

  Show description for 3fc3e2b
  wuxibin89

  authored
  3fc3e2bView commit details

  Copy full SHA for 3fc3e2b

  View code at this pointBrowse repository at this point

• [bugfix] fix: generation script (#542)

  Show description for 79e072f
  nwiad

  authored
  79e072fView commit details

  Copy full SHA for 79e072f

  View code at this pointBrowse repository at this point

• [rollout] feat: support sampling in validation stage (#553)

  Show description for d5de9f4
  PeterSH6

  authored
  d5de9f4View commit details

  Copy full SHA for d5de9f4

  View code at this pointBrowse repository at this point

Commits on Mar 12, 2025

• [misc] add assertion for normalized ppo_mini_batch_size (#552)

  Zeetc

  authored
  39b008dView commit details

  Copy full SHA for 39b008d

  View code at this pointBrowse repository at this point

• [fix] Fix config param issue (#558)

  BearBiscuit05

  authored
  329dcfeView commit details

  Copy full SHA for 329dcfe

  View code at this pointBrowse repository at this point

• refactor: remove custom vllm weight loader and use model.load_weights directly (#543)

  Show description for 6680185
  wuxibin89

  authored
  6680185View commit details

  Copy full SHA for 6680185

  View code at this pointBrowse repository at this point

• Add Math-Verify Support (#545)

  Show description for d4a00ef
  yyDing1

  authored
  d4a00efView commit details

  Copy full SHA for d4a00ef

  View code at this pointBrowse repository at this point

• [Efficiency] feat: remove unnecessary empty_cache (#556)

  Show description for 51a5ff9
  tongyx361

  authored
  51a5ff9View commit details

  Copy full SHA for 51a5ff9

  View code at this pointBrowse repository at this point

Commits on Mar 11, 2025

• [misc] feat: support vllm>0.7 world size 1 generation (#520)

  PeterSH6

  authored
  be2c705View commit details

  Copy full SHA for be2c705

  View code at this pointBrowse repository at this point

Commits on Mar 10, 2025

• recipe: PRIME algorithm (#362)

  Show description for f0e7f9f
  ZefanW

  authored
  f0e7f9fView commit details

  Copy full SHA for f0e7f9f

  View code at this pointBrowse repository at this point

• [bugfix] Fix position embedding processing for Qwen2.5-VL (#527)

  Show description for 830fab2
  

  Yifan-Song793

  and

  songyifan

  authored
  830fab2View commit details

  Copy full SHA for 830fab2

  View code at this pointBrowse repository at this point

• [perf] fix: correct meta weight init error to support hsdp (#508)

  Show description for 872022d
  zhr2001

  authored
  872022dView commit details

  Copy full SHA for 872022d

  View code at this pointBrowse repository at this point

Commits on Mar 8, 2025

• fix `_build_model_optimizer` when role is rollout, whose `optim_config` is None (#322)

  HaoshengZou

  authored
  f8acd90View commit details

  Copy full SHA for f8acd90

  View code at this pointBrowse repository at this point

• feat: support loading reward function from an external file (#452)

  dirtyDan0

  authored
  13a87c7View commit details

  Copy full SHA for 13a87c7

  View code at this pointBrowse repository at this point

Commits on Mar 7, 2025

• Resolve the issue of PRIME getting stuck during math verification. (#469)

  Show description for b742303
  Raf-Chen

  authored
  b742303View commit details

  Copy full SHA for b742303

  View code at this pointBrowse repository at this point

• [misc] feat: make filter long prompt an option (#506)

  Show description for 386cfab
  PeterSH6

  authored
  386cfabView commit details

  Copy full SHA for 386cfab

  View code at this pointBrowse repository at this point

• fix missing raise keyword in NotImplementedError for hdfs loading (#507)

  xffxff

  authored
  fbad52eView commit details

  Copy full SHA for fbad52e

  View code at this pointBrowse repository at this point

• misc: precheck resource pool available to prevent pg hang (#505)

  Show description for 0f0bc5a
  wuxibin89

  authored
  0f0bc5aView commit details

  Copy full SHA for 0f0bc5a

  View code at this pointBrowse repository at this point

• Verl's megatron core_r0.11.0 backend successfully tested with 3D parallelism with multiple bug fixed (#495)

  Show description for 35555d8
  

  ETOgaosion

  and

  uygnef

  authored
  35555d8View commit details

  Copy full SHA for 35555d8

  View code at this pointBrowse repository at this point

• [ckpt] sort pgs by node ip to make RANK consistent across nodes (#500)

  wuxibin89

  authored
  becf7cbView commit details

  Copy full SHA for becf7cb

  View code at this pointBrowse repository at this point

Commits on Mar 6, 2025

• fix: (1) skipped last step (2) redundant validation and logging (#409)

  Show description for 3165d98
  dirtyDan0

  authored
  3165d98View commit details

  Copy full SHA for 3165d98

  View code at this pointBrowse repository at this point

• [misc] feat: add allgather method to dataproto (#497)

  Show description for 0cc2bda
  vermouth1992

  authored
  0cc2bdaView commit details

  Copy full SHA for 0cc2bda

  View code at this pointBrowse repository at this point

• [Hardware] Support AMD (Rocm kernel) (#360)

  yushengsu-thu

  authored
  4a291faView commit details

  Copy full SHA for 4a291fa

  View code at this pointBrowse repository at this point

• [feat] add val_generations_to_log_to_swanlab (#480)

  Show description for 75dedb5
  Zeyi-Lin

  authored
  75dedb5View commit details

  Copy full SHA for 75dedb5

  View code at this pointBrowse repository at this point