Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit history

Commits on May 5, 2025

• fix: update RayPPOTrainer to use dynamic filtering and worker parameters; clarify prompt handling in MathQuestionAnswerDataset

  drproduck

  committed
  7d6e87fView commit details

  Copy full SHA for 7d6e87f

  View code at this pointBrowse repository at this point

Commits on May 4, 2025

• fix: update sample logging logic to handle negative generations_to_log value

  drproduck

  committed
  eb3bebcView commit details

  Copy full SHA for eb3bebc

  View code at this pointBrowse repository at this point

Commits on May 2, 2025

• fix: enhance MathQuestionAnswerDataset to support filtering overlong prompts and add num_workers parameter

  drproduck

  committed
  47dbd13View commit details

  Copy full SHA for 47dbd13

  View code at this pointBrowse repository at this point

Commits on May 1, 2025

• fix: handle None values in validation metrics processing and update prompt usage in RayPPOTrainer

  drproduck

  committed
  3b3a110View commit details

  Copy full SHA for 3b3a110

  View code at this pointBrowse repository at this point

Commits on Apr 25, 2025

• fix: remove unused critic prompt processing in MathQuestionAnswerDataset

  drproduck

  committed
  ed218b3View commit details

  Copy full SHA for ed218b3

  View code at this pointBrowse repository at this point

Commits on Apr 23, 2025

• fix: update parameter name from parquet_files to data_files in MathQuestionAnswerDataset

  drproduck

  committed
  456064bView commit details

  Copy full SHA for 456064b

  View code at this pointBrowse repository at this point

• Merge branch 'main' of https://github.com/drproduck/verl into main

  drproduck

  committed
  63c0d9dView commit details

  Copy full SHA for 63c0d9d

  View code at this pointBrowse repository at this point

• add critic prompt

  drproduck

  committed
  a00db98View commit details

  Copy full SHA for a00db98

  View code at this pointBrowse repository at this point

• fix: correct prompt handling in MathQuestionAnswerDataset

  drproduck

  committed
  b13cafaView commit details

  Copy full SHA for b13cafa

  View code at this pointBrowse repository at this point

• refactor: support loading other data_file types

  drproduck

  committed
  23d6284View commit details

  Copy full SHA for 23d6284

  View code at this pointBrowse repository at this point

Commits on Apr 9, 2025

• chore: import additional libraries in dataset and rollout files

  Show description for 0503ad9
  drproduck

  committed
  0503ad9View commit details

  Copy full SHA for 0503ad9

  View code at this pointBrowse repository at this point

Commits on Apr 8, 2025

• - Removed duplicate `val_before_train` setting from `ppo_trainer.yaml`

  drproduck

  committed
  50c1107View commit details

  Copy full SHA for 50c1107

  View code at this pointBrowse repository at this point

• Merge branch 'main' of https://github.com/drproduck/verl into main

  drproduck

  committed
  e2d06ddView commit details

  Copy full SHA for e2d06dd

  View code at this pointBrowse repository at this point

• Merge branch 'main' of https://github.com/volcengine/verl into main

  drproduck

  committed
  15add0dView commit details

  Copy full SHA for 15add0d

  View code at this pointBrowse repository at this point

• fix: reward_fn_key for PRIME (#975)

  tongyx361

  authored
  1a42f14View commit details

  Copy full SHA for 1a42f14

  View code at this pointBrowse repository at this point

• fix: return list from bootstrap_metric (#969)

  Show description for 6433fd4
  tongyx361

  authored
  6433fd4View commit details

  Copy full SHA for 6433fd4

  View code at this pointBrowse repository at this point

• fix: optim.warmup_style do not take effect (#418) (#959)

  Show description for fd0eba0
  czczup

  authored
  fd0eba0View commit details

  Copy full SHA for fd0eba0

  View code at this pointBrowse repository at this point

• Merge branch 'main' of https://github.com/drproduck/verl into main

  drproduck

  committed
  ec382fbView commit details

  Copy full SHA for ec382fb

  View code at this pointBrowse repository at this point

• refactor: clean up unused assertions and reorganize metric logging in RayPPOTrainer

  Show description for dde687a
  drproduck

  committed
  dde687aView commit details

  Copy full SHA for dde687a

  View code at this pointBrowse repository at this point

Commits on Apr 7, 2025

• [distributed] enhancement: Make `register_center` named actor waiting time configurable & providing better error info (#947)

  Show description for c87e9f6
  

  hongpeng-guo

  and

  Copilot

  authored
  c87e9f6View commit details

  Copy full SHA for c87e9f6

  View code at this pointBrowse repository at this point

• [megatron] feat: add gradient checkpointing in megatron backend (#944)

  Show description for d13434f
  ccclyu

  authored
  d13434fView commit details

  Copy full SHA for d13434f

  View code at this pointBrowse repository at this point

• feat: Batch Rewards (#871)

  mertunsall

  authored
  82cbc43View commit details

  Copy full SHA for 82cbc43

  View code at this pointBrowse repository at this point

Commits on Apr 6, 2025

• [sglang] feat: enhence sglang_rollout to handle image input (#824)

  Show description for 6efa018
  

  3 peopleauthored
  6efa018View commit details

  Copy full SHA for 6efa018

  View code at this pointBrowse repository at this point

• [vllm] feat: enable FSDP and vLLM 0.8.2 to support DPSK v3 training. (#923)

  Show description for f8d1973
  BearBiscuit05

  authored
  f8d1973View commit details

  Copy full SHA for f8d1973

  View code at this pointBrowse repository at this point

• [log] fix: validation metrics of reward and maj voting (#927)

  Show description for d2c6064
  tongyx361

  authored
  d2c6064View commit details

  Copy full SHA for d2c6064

  View code at this pointBrowse repository at this point

• [rf++] style: eos_mask to response_mask for reinforce++-baseline method (#938)

  Show description for 7afa6c6
  none0663

  authored
  7afa6c6View commit details

  Copy full SHA for 7afa6c6

  View code at this pointBrowse repository at this point

• prompt: Fix computer_score in fsdp_workers.py (#629)

  Show description for 15263cb
  

  Django-Jiang

  and

  eric-haibin-lin

  authored
  15263cbView commit details

  Copy full SHA for 15263cb

  View code at this pointBrowse repository at this point

• fix: seq-mean-token-sum loss (#931)

  tongyx361

  authored
  4d722b1View commit details

  Copy full SHA for 4d722b1

  View code at this pointBrowse repository at this point

• [sglang] feat: SGLang rollout multinode support (#915)

  Show description for 7fc8330
  

  3 peopleauthored
  7fc8330View commit details

  Copy full SHA for 7fc8330

  View code at this pointBrowse repository at this point

• Merge branch 'main' of https://github.com/volcengine/verl into main

  drproduck

  committed
  3e0f003View commit details

  Copy full SHA for 3e0f003

  View code at this pointBrowse repository at this point

• [vllm] fix: skip vllm initialization with weight loading (#922)

  Show description for de46048
  BearBiscuit05

  authored
  de46048View commit details

  Copy full SHA for de46048

  View code at this pointBrowse repository at this point

Commits on Apr 5, 2025

• [math-verify] fix: TimeoutException (#929)

  tongyx361

  authored
  8447937View commit details

  Copy full SHA for 8447937

  View code at this pointBrowse repository at this point

• Feat: support REINFORCE++-baseline and add script for REINFORCE++ (#908)

  Show description for 30259d2
  hijkzzz

  authored
  30259d2View commit details

  Copy full SHA for 30259d2

  View code at this pointBrowse repository at this point

Commits on Apr 4, 2025

• feat: Add multi-turn SFT support (#195)

  xingyaoww

  authored
  fb03941View commit details

  Copy full SHA for fb03941

  View code at this pointBrowse repository at this point

• [dataset] refactor: remove unused filter_prompts parameter from RLHFDataset (#889)

  Show description for 0407cad
  0x404

  authored
  0407cadView commit details

  Copy full SHA for 0407cad

  View code at this pointBrowse repository at this point