fix lightllm vit triton

[shihaobai]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello @shihaobai, 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 critical numerical stability issue within the Triton-optimized FlashAttention kernel used for Vision Transformers in LightLLM. The changes involve correcting the initialization and an intermediate calculation within the attention mechanism's log-sum-exp trick, which is vital for maintaining accuracy and stability during model inference.

Highlights

• FlashAttention Numerical Stability: Corrected the initialization of the l_i variable, which accumulates log-sum-exp values, from 0.0 to -float("inf") within the _fwd_kernel function. This is a standard practice for ensuring numerical stability in log-sum-exp computations, preventing potential underflow issues when probabilities are very small.
• Log-Sum-Exp Maximum Correction: Modified the calculation of m_ij to use m_i (the current maximum logit) instead of l_i (the log-sum-exp accumulator) when determining the maximum value for the log-sum-exp trick. This ensures the correct maximum is used for scaling, which is crucial for preventing numerical overflow or underflow during the exponentiation step in attention calculations.

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

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 addresses two critical bugs in the Triton kernel for flash attention (flashattention_nopad.py). The fixes correct the initialization of the log-sum-exp accumulator and the update logic for the running maximum value. These changes are essential for the correctness of the attention mechanism in Vision Transformer models.