verl-project/verl

Additional memory optimization features

Open

#144 建立於 2025年1月27日

在 GitHub 查看
 (4 留言) (3 反應) (0 負責人)Python (21,533 star) (3,940 fork)auto 404
call for contributionenhancementgood first issue

描述

  • Activation offloading (see implementation here)
  • Fusing optimizer step into backward pass (see implementation here)
  • Utilize full_shard reshard_after_forward (see here). I wasn't 100% sure if I could see this already implemented in veRL.

These optimizations largely trade off decreased peak memory useage for additional compute, so may only be useful for training larger models, and in GPU-constrained settings.

貢獻者指南

技術棧
pythonpytorch
領域
machine learningperformance
議題類型
feature
難度面向新貢獻者的預計實作難度，1 表示很小改動，5 表示專家級工作。
4
預計時間有經驗貢獻者完成調查、實作、測試並準備 pull request 的粗略時間範圍。
over 1 week
活動狀態議題目前的可參與程度：新鮮、活躍、陳舊、阻塞或等待維護者輸入。
fresh
清晰度議題是否清楚說明預期改動、驗收標準和下一步。
mostly clear
前置要求
PythonPyTorchFSDP
新手友善度1-100 的估計分數，表示該議題對首次貢獻者的友善程度。
15
研究方向
Explore the provided links to activation offloading and fused optimizer step implementations from torchtune (torchtune/training/ activation offloading.py and torchtune/training/memory.py). Study the FSDP2 API differences in torchtitan (docs/fsdp.md). Then, survey the verI codebase to identify where similar memory optimizations can be integrated, focusing on the training loop and model sharding. Consider discussing with maintainers on which optimization is most needed first.