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.

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

[Auto Parallel] Sharding Optimization:Partition Algorithm & Stage2 Parameter Bucket communication #47180

Merged

Conversation

JZ-LIANG
Copy link
Contributor

@JZ-LIANG JZ-LIANG commented Oct 19, 2022

PR types

Performance optimization

PR changes

Others

Describe

add a new parameter partition algorithm for sharding which will favor bucket communication.
Bucket communication for stage2 parameter.

GPT3-6.7B Before (token/s) After (token/s) Improv
N1C8 47185 49676 5.3%
N2C16 32768 33816 3.2%

@paddle-bot
Copy link

paddle-bot bot commented Oct 19, 2022

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.

@JZ-LIANG JZ-LIANG changed the title partition param by order [Auto Parallel] Sharding Pass Partition param by construction order Oct 19, 2022
@JZ-LIANG JZ-LIANG changed the title [Auto Parallel] Sharding Pass Partition param by construction order [Auto Parallel] Sharding:Partition param by construction order Oct 25, 2022
@JZ-LIANG JZ-LIANG changed the title [Auto Parallel] Sharding:Partition param by construction order [Auto Parallel] Sharding Optimization:Partition Algorithm & Stage2 Parameter Bucket communication Nov 7, 2022
Copy link
Contributor

@aoyulong aoyulong left a comment

Choose a reason for hiding this comment

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

LGTM

@JZ-LIANG JZ-LIANG merged commit e5eb3f5 into PaddlePaddle:develop Nov 8, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants