verl-project/verl

[FR] Tests for gradient accumulation

クローズ

#907 opened on 2025/04/03

 (0 件のコメント) (0 件のリアクション) (0 人の担当者)Python (4,427 件のフォーク)auto 404
enhancementgood first issue

Repository metrics

Stars
 (23,038 個のスター)
PR merge metrics
 (平均マージ 3d 6h) (30d で 125 merged PRs)

説明

Summary

Promise from #906.

Add tests to assert the losses on the same mini-batch are equal

  • between w/o & w/ gradient accumulation
  • using all the different loss aggregation modes

Plan

We might went to test end-to-end to avoid potential implication by any other changes.

Plan Pro
A) Inherit a new Trainer class to test Standalone without affecting the main config
B) Add a config term and modify update_* to test 1) Minimal code modification 2) Compatibility to any Trainer

コントリビューターガイド