[bugfix] fix sp & loss_scale #5497
Conversation
There was a problem hiding this comment.
Summary of Changes
Hello @Jintao-Huang, 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 addresses a bug in the compute_loss function within the training module. The fix ensures that certain loss calculations, specifically those involving per_token_loss_func and loss_scale, are only applied when sequence parallelism is not active, preventing potential issues in distributed training environments.
Highlights
- Loss Computation Logic: Modified the
compute_lossmethod to conditionally applyper_token_loss_funcandloss_scaleonly whenself.template.sequence_parallel_sizeis equal to 1.
Using Gemini Code Assist
The 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 in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| Feature | Command | Description |
|---|---|---|
| Code Review | /gemini review |
Performs a code review for the current pull request in its current state. |
| Pull Request Summary | /gemini summary |
Provides a summary of the current pull request in its current state. |
| Comment | @gemini-code-assist | Responds in comments when explicitly tagged, both in issue comments and review comments. |
| Help | /gemini help |
Displays a list of available commands. |
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 .gemini/ folder in the base of the repository. Detailed instructions can be found here.
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
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
There was a problem hiding this comment.
Code Review
This pull request fixes an issue with loss calculation when sequence parallelism (SP) is enabled. The change correctly disables the non-SP loss calculation path when SP is active, which avoids incorrect or redundant computations. The change is logical, but I have added a comment regarding a potential fragility this introduces in the code.
| @@ -359,7 +359,7 @@ def compute_loss(self, model, inputs, return_outputs=False, num_items_in_batch=N | |||
| loss = outputs['loss'] if isinstance(outputs, dict) else outputs[0] | |||
| else: | |||
| outputs.loss = None | |||
| if self.args.enable_dft_loss or loss_scale is not None: | |||
| if self.template.sequence_parallel_size == 1 and (self.args.enable_dft_loss or loss_scale is not None): | |||
There was a problem hiding this comment.
This change correctly prevents this block from executing when sequence parallelism is enabled, which is the right thing to do as the loss is handled by compute_loss_func in that case.
However, this makes the code more fragile. If sequence_parallel_size > 1 and for some reason compute_loss_func is not set, outputs.loss will be None, leading to a NoneType error at line 382. This creates a strong, implicit dependency on compute_loss_func being set by sequence parallel implementations.
It would be more robust to handle this possibility explicitly, for example by raising an informative error if sequence_parallel_size > 1 but compute_loss_func is None.
No description provided.