fix: add trtllm-allreduce-fusion api notes and fix memory error#1229
fix: add trtllm-allreduce-fusion api notes and fix memory error#1229yzh119 merged 7 commits intoflashinfer-ai:mainfrom
Conversation
There was a problem hiding this comment.
Summary of Changes
Hello @yyihuang, 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 focuses on improving the robustness and clarity of the trtllm_allreduce_fusion API. It enhances the API's documentation to provide clearer guidance on its usage and implements a critical runtime check to prevent a specific memory error, thereby making the API more reliable for downstream integrations.
Highlights
- API Documentation Enhancement: Added detailed usage notes to the
trtllm_allreduce_fusionAPI, specifically guiding when theuse_oneshotparameter should be enabled or disabled based on strategy preference, latency mode, and sequence length, to aid integration with systems like vLLM/SGL. - Memory Error Prevention: Introduced a runtime assertion within the
trtllm_allreduce_fusionfunction to ensure that the inputsequence lengthis greater than theworld_sizewhenuse_oneshotis disabled. This change directly addresses a reported memory error, likely related to issue #1223.
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 configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
yzh119
left a comment
There was a problem hiding this comment.
LGTM, thank you for the quick fix!
📌 Description
We add some notes on each api to help integration to vllm/sgl.
And we try to fix the error in issues.
To avoid workspace size overflow, the max lamport communication size MAX_COMM_SIZE (computed as hidden * max_token) should be less than round_down(INT_32_MAX, 2MB). For any larger size, it would be rewritten to this value with a warning "warning: lamport_comm_size 2147483648 is greater than MAX_COMM_SIZE 2145386496, set to MAX_COMM_SIZE".
If the actual lamport communication size (computed as hidden * token_num) exceeds this MAX_COMM_SIZE, always set use_oneshot to be False.
Otherwise, you could set use_oneshot on your preference; and for the min-latency case, set it to be (token_num <= 128).
🔍 Related Issues
#1223
sgl-project/sglang#7621
🚀 Pull Request Checklist
Thank you for contributing to FlashInfer! Before we review your pull request, please make sure the following items are complete.
✅ Pre-commit Checks
pre-commitby runningpip install pre-commit(or used your preferred method).pre-commit install.pre-commit run --all-filesand fixed any reported issues.🧪 Tests
unittest, etc.).Reviewer Notes