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

Add Pipeline Parallel for PPO training and support generation with InferenceModel #7953

Merged
merged 56 commits into from
Jun 13, 2024

Conversation

guoshengCS
Copy link
Contributor

@guoshengCS guoshengCS commented Feb 2, 2024

PR types

New features

PR changes

Others

Description

Add Pipeline Parallel for PPO training and support generation with InferenceModel

Copy link

codecov bot commented Feb 2, 2024

Codecov Report

Attention: Patch coverage is 30.55556% with 25 lines in your changes missing coverage. Please review.

Project coverage is 54.41%. Comparing base (b36b6a0) to head (f1e66f2).
Report is 290 commits behind head on develop.

Files with missing lines Patch % Lines
paddlenlp/utils/nested.py 11.11% 16 Missing ⚠️
paddlenlp/transformers/llama/modeling_pp.py 28.57% 5 Missing ⚠️
paddlenlp/generation/utils.py 33.33% 2 Missing ⚠️
paddlenlp/trainer/plugins/timer.py 0.00% 1 Missing ⚠️
paddlenlp/transformers/model_utils.py 0.00% 1 Missing ⚠️
Additional details and impacted files
@@             Coverage Diff             @@
##           develop    #7953      +/-   ##
===========================================
+ Coverage    54.29%   54.41%   +0.12%     
===========================================
  Files          617      632      +15     
  Lines        96339    99476    +3137     
===========================================
+ Hits         52310    54134    +1824     
- Misses       44029    45342    +1313     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

@guoshengCS guoshengCS requested review from ZHUI and wawltor February 22, 2024 07:59
@guoshengCS guoshengCS changed the title Add Pipeline Parallel for PPO training. Add Pipeline Parallel for PPO training and support generation with InferenceModel Mar 13, 2024
@gongel gongel marked this pull request as ready for review June 11, 2024 03:19
@CLAassistant
Copy link

CLAassistant commented Jun 12, 2024

CLA assistant check
All committers have signed the CLA.

# and sampling, and then broadcast to avoid broadcast logits.
if hasattr(self, "pp_group"):
paddle.distributed.broadcast(
next_tokens, src=self.pp_group.ranks[0], group=self.pp_group # use rank 0 for same seed to check
Copy link
Member

Choose a reason for hiding this comment

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

我看CI这里,self.pp_group有可能为None,导致.ranks[0]报错

Copy link
Collaborator

@wawltor wawltor left a comment

Choose a reason for hiding this comment

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

LGTM

@wawltor wawltor merged commit bf6d4e7 into PaddlePaddle:develop Jun 13, 2024
7 of 12 checks passed
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.

5 participants