-
Notifications
You must be signed in to change notification settings - Fork 55
Add new sync style TRAINER_DRIVEN.
#494
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
Conversation
2. Refactor explorer and fix the issue of inconsistent behavior between NCCL and checkpoint modes in the case of `DYNAMIC_BY_EXPLORER`.
Summary of ChangesHello @chenyushuo, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request enhances the system's distributed training capabilities by introducing a new synchronization style, Highlights
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Code Review
This pull request introduces a new synchronization style DYNAMIC_BY_TRAINER and performs a significant refactoring of the synchronization logic between the explorer and trainer. The changes simplify the state machine by removing the WAITING_SYNC status and centralizing state transitions within the Synchronizer. A crucial fix is included for DYNAMIC_BY_EXPLORER mode to prevent the explorer from hanging when the trainer has no new model to sync. The tests have also been substantially refactored for better readability and maintainability, which is a great improvement.
My main feedback points are:
- A potential semantic change in the
DYNAMIC_BY_EXPLORERsync style that makes it behave like a fixed-step schedule. - A suggestion to simplify the logic in a test monkey patch to improve clarity and remove a redundant call.
Overall, this is a solid refactoring with valuable improvements.
|
/unittest-all |
2. Bug fix and refactor in checkpoint load. 3. Fix when trainer use one gpu and memory sync method.
|
/unittest-all |
Summary
Failed Tests
Skipped
Tests
Github Test Reporter by CTRF 💚 |
|
/unittest-all |
Summary
Skipped
Tests
Github Test Reporter by CTRF 💚 |
…iner` to `trainer_driven` in SyncStyle
DYNAMIC_BY_TRAINER.TRAINER_DRIVEN.
|
/unittest-all |
Summary
Failed Tests
Skipped
Tests
Github Test Reporter by CTRF 💚 |
|
/unittest-all |
95afdeb to
c8f4a58
Compare
|
/unittest-all |
Summary
Failed Tests
Skipped
Tests
Github Test Reporter by CTRF 💚 |
|
/unittest-module-cli |
Summary
Failed Tests
Tests
Github Test Reporter by CTRF 💚 |
|
/unittest-module-cli |
Summary
Tests
Github Test Reporter by CTRF 💚 |
Description
TRAINER_DRIVEN.DYNAMIC_BY_EXPLORERtoEXPLORER_DRIVENwhich maintains forward compatibility.EXPLORER_DRIVEN.chat_templatein vllm openai API server.Checklist
Please check the following items before code is ready to be reviewed.