fastrope fix for zero strided tensors

[f14-bertolotti]

Here's the fix for #3781

Briefly, in some scenarios, zero strided tensors may be feeded to the backward or forward pass of the fast rope kernel.

In this scenarios, the tensor materialization was not handled properly. This PR fixes this issue.

[gemini-code-assist]

Summary of Changes

Hello @f14-bertolotti, 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 issue where zero-strided tensors were not correctly processed within the fast rope kernel, specifically impacting tensor materialization during both forward and backward computations. The changes ensure that tensor contiguity checks are performed using the correct method and that subsequent stride calculations accurately reflect the state of potentially cloned tensors, thereby preventing erroneous behavior when non-contiguous inputs are encountered.

Highlights

• Syntax Correction: Updated the is_contiguous attribute access to is_contiguous() method calls for PyTorch tensors in both the forward and backward passes of the rope embedding kernel. This ensures correct evaluation of tensor contiguity.
• Corrected Stride Calculation for Zero-Strided Tensors: In the backward pass, the logic for creating dQ_out and dK_out (which are clones if the input tensors are not contiguous) was moved before the stride calculations. This ensures that the stride values are correctly derived from the potentially materialized dQ_out and dK_out tensors, addressing issues with zero-strided tensors not being handled properly.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

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 refactors the rope_embedding.py kernel by correcting the is_contiguous attribute to a method call is_contiguous() for Q, K, dQ, and dK tensors. Additionally, in the backward pass, the assignments for dQ_out and dK_out are moved to occur before the stride calculations, ensuring that the strides are computed from the potentially cloned, contiguous tensors rather than the original input tensors.

[danielhanchen]

Nice work @f14-bertolotti and thanks!