bugfix: Fix precision issue with Triton operator token_att_fwd

[WuSiYu]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello @WuSiYu, 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 resolves a numerical precision problem within the Triton token_att_fwd operator. By explicitly casting the intermediate sum of attention values to float32 during computation, the change prevents potential inaccuracies that could arise from implicit type handling, thereby enhancing the stability and correctness of the model's attention mechanism.

Highlights

• Precision Fix: Addressed a precision issue in the token_att_fwd Triton operator by explicitly setting the dtype of the tl.sum operation to tl.float32.
• Triton Kernel Update: Modified the token_attention_nopad_att1.py kernel to ensure attention value calculations maintain float32 precision throughout the summation process.

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 by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

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 pull request 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

1. 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. ↩

[gemini-code-assist]

Code Review

This pull request correctly addresses a potential precision issue in the _fwd_kernel_token_att1 Triton kernel. By changing tl.sum(...) followed by .to(tl.float32) to tl.sum(..., dtype=tl.float32), the summation is performed using a float32 accumulator. This prevents precision loss when summing many low-precision floating-point numbers and improves numerical stability. The change is also more efficient.

This is an excellent fix. I suggest auditing other Triton kernels in the repository for similar patterns. For instance, _fwd_kernel_token_att1_int8 in the same file and kernels in lightllm/models/llama/triton_kernel/token_attention_nopad_reduceV.py seem to have summations over low-precision types that could also benefit from using a float32 accumulator to prevent potential silent precision loss.