Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[XPU] avoid pre-allocating gm buffer #60387

Merged
merged 1 commit into from
Dec 29, 2023
Merged

[XPU] avoid pre-allocating gm buffer #60387

merged 1 commit into from
Dec 29, 2023

Conversation

dynamicheart
Copy link
Contributor

@dynamicheart dynamicheart commented Dec 27, 2023 •
edited
Loading

PR types

Performance optimization

PR changes

Others

Description

  • Avoid pre-allocating XPU gm buffer(default is 128MB) in ProcessGroupBKCL communication XPUContexts and DataLoader XPUContexts as they does not use this pre-allocated GM buffers.
  • In hybrid-parallel distributed training, multiple process groups with different XPUContexts will be created for different parallel methods. See topology.py
  • Multi-processes dataloader will also create multiple XPUContexts.

Related PR: #60260

@paddle-bot Paddle-bot
Copy link

paddle-bot bot commented Dec 27, 2023

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

@dynamicheart
Copy link
Contributor Author

@XiaociZhang 麻烦帮忙review下将ProcessGroupBKCL的comm_ctx的GM buffer释放是否合理。

@XiaociZhang
Copy link
Contributor

LGTM on process group

zhangyk0314
zhangyk0314 approved these changes Dec 29, 2023
View reviewed changes
Copy link
Contributor

@zhangyk0314 zhangyk0314 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@QingshuChen QingshuChen merged commit c15a6d3 into PaddlePaddle:develop Dec 29, 2023
29 checks passed
Wanglongzhi2001 pushed a commit to Wanglongzhi2001/Paddle that referenced this pull request Jan 7, 2024
@dynamicheart
[XPU] avoid pre-allocating gm buffer (PaddlePaddle#60387)
00305e7
@houj04 houj04 added the XPU label Sep 4, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@zhangyk0314 zhangyk0314 zhangyk0314 approved these changes

@cqulilujia cqulilujia cqulilujia approved these changes

@QingshuChen QingshuChen QingshuChen approved these changes

Assignees
No one assigned
Labels
XPU
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
6 participants
@dynamicheart @XiaociZhang @QingshuChen @cqulilujia @zhangyk0314 @houj04