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

[hybrid performance] all reduce fusion for sharding #34480

Merged
merged 1 commit into from
Jul 30, 2021

Conversation

FeixLiu
Copy link
Contributor

@FeixLiu FeixLiu commented Jul 29, 2021

PR types

New features

PR changes

Others

Describe

allreduce fuse supports for sharding_optimizer

Throughput test

Using GPT model, 8 * V100, fuse_grad_in_size=32MB

dp=4 sharding=2

No Fuse Fused Gain
throughput 135503 tokens/s 138892 tokens/s +2.7%
allreduce number 57 15 -73%

Loss curve

dp_hybrid_sharding
Screen Shot 2021-07-29 at 5 39 13 PM

dp_pp_hybrid_sharding
Screen Shot 2021-07-29 at 8 36 11 PM

@paddle-bot-old
Copy link

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

Copy link
Contributor

@wangxicoding wangxicoding left a comment

Choose a reason for hiding this comment

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

LGTM

@wangxicoding wangxicoding merged commit 423ea97 into PaddlePaddle:develop Jul 30, 2021
@FeixLiu FeixLiu deleted the allreduce_fuse_sharding branch July 30, 2021 02:06
@FeixLiu FeixLiu changed the title all reduce fusion for shardinug [hybrid performance] all reduce fusion for shardinug Oct 11, 2021
@FeixLiu FeixLiu changed the title [hybrid performance] all reduce fusion for shardinug [hybrid performance] all reduce fusion for sharding Oct 11, 2021
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