feat: add lse return on trtllm-gen attention #1566
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 @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 introduces the capability to return the Log-Sum-Exp (LSE) value from the trtllm-gen attention kernels. This enhancement provides access to intermediate attention computation results, which can be crucial for debugging, advanced analysis, or implementing specific loss functions that rely on these values. The changes span both the CUDA kernel implementations and their Python bindings, ensuring that LSE can be optionally returned from both prefill and decode operations.
Highlights
- CUDA Kernel Update: The core
trtllm_paged_attention_launcherCUDA kernel now accepts anlse(Log-Sum-Exp) output parameter, enabling the direct return of these intermediate attention computation values. - Python API Enhancement: The Python API functions, specifically
trtllm_batch_decode_with_kv_cache,trtllm_batch_context_with_kv_cache, and their MLA (Multi-Layer Attention) counterparts, have been enhanced to optionally return thelsetensor. - Automatic Tensor Allocation: Automatic allocation of the
lsetensor is now supported when thereturn_lseflag is enabled and nolsetensor is explicitly provided, simplifying usage. - Comprehensive Test Coverage: Unit tests have been updated across various attention functions to include verification of the returned
lsevalues against established reference implementations, ensuring correctness and numerical stability.
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. ↩
|
Warning Gemini encountered an error creating the review. You can try again by commenting |
…n attention (#1584) <!-- .github/pull_request_template.md --> ## 📌 Description **workspace_buffer arrangement** - on main branch paged_attention: multiCtasKvCounter or semaphores | multiCtasKvScratch ragged_attention: softmax | multiCtasKvCounter or semaphores | multiCtasKvScratch - on PR branch: softmax (optional) | multiCtasKvScratch | multiCtasKvCounter or semaphores (last 8MB of 128 MB) The range of semaphores must be fixed across multiple execution, since we are not clearing the buffer by zeros explicitly any more. related PR: #1463 And #1566 (WIP) depends on this. ## 🔍 Related Issues <!-- Link any related issues here --> ## 🚀 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 - [x] I have installed `pre-commit` by running `pip install pre-commit` (or used your preferred method). - [x] I have installed the hooks with `pre-commit install`. - [x] I have run the hooks manually with `pre-commit run --all-files` and fixed any reported issues. > If you are unsure about how to set up `pre-commit`, see [the pre-commit documentation](https://pre-commit.com/). ## 🧪 Tests - [x] Tests have been added or updated as needed. - [x] All tests are passing (`unittest`, etc.). ## Reviewer Notes <!-- Optional: anything you'd like reviewers to focus on, concerns, etc. -->
|
Can we merge this before v0.3.0 @yzh119 |
|
91e3b83 this commit should fix fp8. The fundamental reason is because trtllm kernels use internal scale for rowsum/rowmax which might not align with provided bmm scale. For fp8, the reason we need to |
4 participants
📌 Description
🔍 Related Issues
sgl-project/sglang#9471
🚀 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