fix: remove attention mask shift & add pe shift

[Liyuhui-12]

Motivation

The implementation of EAGLE-3 uses an unnecessary attention mask shift and is missing the position encoding shift.

Modifications

Remove attention mask shift & Add pe shift

Related Issues

Accuracy Test

Benchmark & Profiling

Checklist

• Format your code according to the Code Formatting with Pre-Commit.
• Add unit tests as outlined in the Running Unit Tests.
• Update documentation / docstrings / example tutorials as needed, according to Writing Documentation.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed, according to Benchmark and Profiling and Accuracy Results.
• For reviewers: If you haven't made any contributions to this PR and are only assisting with merging the main branch, please remove yourself as a co-author when merging the PR.
• Please feel free to join our Slack channel at https://sgl-fru7574.slack.com/archives/C09784E3EN6 to discuss your PR.

[gemini-code-assist]

Summary of Changes

Hello @Liyuhui-12, 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 refines the EAGLE-3 implementation by correcting how attention masks and position encodings are handled. It removes an erroneous and unnecessary attention mask shift while introducing a crucial, previously missing, position encoding shift. These adjustments aim to improve the correctness and alignment of the model's internal computations.

Highlights

• Removed Unnecessary Attention Mask Shift: The explicit attention mask shifting logic within the forward method of eagle3.py for the sdpa attention backend has been removed. This change addresses an unnecessary manipulation of the attention mask that was present in the original implementation.
• Added Position Encoding Shift: A line position_ids += 1 has been introduced in the forward method of eagle3.py. This addition correctly applies a position encoding shift, which was previously missing and is crucial for the proper functioning of the EAGLE-3 implementation.

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 refactors the Test-Time Training (TTT) loop logic by replacing a manual attention mask shift with a more appropriate position ID increment. This change simplifies the code and fixes an issue where position encodings were not being shifted correctly. The code is now cleaner and more aligned with standard transformer practices. I've added one comment regarding the need to update unit tests to reflect this logic change.

[gemini-code-assist]

This change to use position_ids += 1 is a great improvement for handling the position encoding shift during the TTT loop. It's cleaner and more correct than manipulating the attention mask directly.

However, this logic change should be reflected in the unit tests. I noticed that tests/test_utils/test_flex_attention.py seems to simulate the TTT loop with the old logic of shifting the attention mask. To ensure the correctness of this fix and prevent future regressions, could you please update the relevant unit tests to use position_ids incrementing instead?