speed up topk #86
speed up topk #86
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Summary of Changes
Hello @RamezesDong, 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 aims to improve the performance of the topk operation within the MoE layers by consolidating and compiling the post-processing logic using torch.compile. This change is expected to make the post-processing steps more efficient.
Highlights
- Performance Optimization: Introduced a new compiled function
_biased_grouped_topk_postprocessto combine and optimize post-processing steps in thebiased_grouped_topkfunction. - Code Refactoring: Refactored the
biased_grouped_topkfunction to call the new compiled post-processing function, replacing separate calls totopk_ids_logical_to_physicaland_mask_topk_ids_padded_regionwhen post-processing is required.
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 is currently in preview and 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 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 configureGemini 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
The pull request introduces a new function _biased_grouped_topk_postprocess and modifies the biased_grouped_topk function to improve the speed of topk operations. The changes involve using torch.compile to optimize the post-processing steps and refactoring the conditional logic for applying these steps.
| @torch.compile(dynamic=True, backend=get_compiler_backend()) | ||
| def _biased_grouped_topk_postprocess( | ||
| topk_ids, expert_location_dispatch_info, num_token_non_padded | ||
| ): | ||
| topk_ids = topk_ids_logical_to_physical(topk_ids, expert_location_dispatch_info) | ||
| _mask_topk_ids_padded_region(topk_ids, num_token_non_padded) | ||
| return topk_ids |
There was a problem hiding this comment.
Consider adding a docstring to _biased_grouped_topk_postprocess to explain its purpose and parameters.
| @torch.compile(dynamic=True, backend=get_compiler_backend()) | |
| def _biased_grouped_topk_postprocess( | |
| topk_ids, expert_location_dispatch_info, num_token_non_padded | |
| ): | |
| topk_ids = topk_ids_logical_to_physical(topk_ids, expert_location_dispatch_info) | |
| _mask_topk_ids_padded_region(topk_ids, num_token_non_padded) | |
| return topk_ids | |
| def _biased_grouped_topk_postprocess( | |
| topk_ids, expert_location_dispatch_info, num_token_non_padded | |
| ): | |
| topk_ids = topk_ids_logical_to_physical(topk_ids, expert_location_dispatch_info) | |
| _mask_topk_ids_padded_region(topk_ids, num_token_non_padded) | |
| return topk_ids |
| # TODO merge into kernel | ||
| if (expert_location_dispatch_info is not None) or ( | ||
| num_token_non_padded is not None | ||
| ): | ||
| topk_ids = _biased_grouped_topk_postprocess( | ||
| topk_ids, expert_location_dispatch_info, num_token_non_padded | ||
| ) | ||
| return topk_weights, topk_ids |
There was a problem hiding this comment.
The conditional check (expert_location_dispatch_info is not None) or (num_token_non_padded is not None) is repeated. Consider extracting this condition into a variable for better readability and maintainability.
| # TODO merge into kernel | |
| if (expert_location_dispatch_info is not None) or ( | |
| num_token_non_padded is not None | |
| ): | |
| topk_ids = _biased_grouped_topk_postprocess( | |
| topk_ids, expert_location_dispatch_info, num_token_non_padded | |
| ) | |
| return topk_weights, topk_ids | |
| topk_weights, topk_ids = moe_fused_gate( | |
| gating_output, | |
| correction_bias, | |
| num_expert_group, | |
| topk_group, | |
| topk, | |
| n_share_experts_fusion, | |
| routed_scaling_factor, | |
| ) | |
| # TODO merge into kernel | |
| condition = (expert_location_dispatch_info is not None) or (num_token_non_padded is not None) | |
| if condition: | |
| topk_ids = _biased_grouped_topk_postprocess( | |
| topk_ids, expert_location_dispatch_info, num_token_non_padded | |
| ) | |
| return topk_weights, topk_ids |
Motivation
Modifications
Checklist