[DP] fix the compatibility issue between DP attention and --attention-backend triton
#8723
Conversation
There was a problem hiding this comment.
Summary of Changes
Hello @ch-wan, 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 compatibility issue between Data Parallel attention and the Triton attention backend. It refines how batch sizes and sequence length sums are managed during the forward pass, particularly in decode mode, to ensure the attention mechanism receives correctly sized inputs while preserving the original batch context for subsequent processing steps.
Highlights
- Compatibility Fix for Data Parallel Attention: Modified the
prepare_mlp_sync_batchmethod to correctly handlebatch_sizeandseq_lens_sumduring decode operations when using Data Parallel (DP) attention with the Triton backend. This involves storing the original batch size (raw_bs) and adjusting the currentbatch_sizeto reflect the number of tokens processed by the specific DP rank (num_tokens). Additionally,seq_lens_sumis now padded to account for the adjusted batch size. - Post-processing Batch Size Correction: Updated the
post_forward_mlp_sync_batchmethod to retrieve the original batch size for post-processing. It now uses the previously storedraw_bsattribute, falling back to the currentself.batch_sizeifraw_bsis not set, ensuring that subsequent operations use the correct global batch context.
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 or fill out our survey to provide feedback.
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 addresses a compatibility issue for DP attention with the Triton backend. The changes involve adjusting the batch size for decode steps in a DP setting and correctly updating seq_lens_sum when padding is applied. The logic seems correct. I have a few suggestions to improve code style and maintainability.
…n-backend triton` (#8723)
…n-backend triton` (#8723)
Motivation
Modifications
Accuracy Test
Benchmark & Profiling
Checklist