Bugfix: Fix Type consistency for KV indices in SWARadixCache

[hzh0425]

Motivation

CUDA device-side assertion triggered in update_sliding_window_buffer when using SWA with radix cache:

Root Cause, Type mismatch in the data pipeline:
KV indices were cloned and stored in radix tree as int32
window_kv_indices is created as int64
Type inconsistency in Triton kernel operations causes invalid memory access

Solution:
Explicitly convert page_aligned_kv_indices to torch.int64 before storing in radix tree

Co-Author-by: @yizhang2077

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.

[gemini-code-assist]

Summary of Changes

Hello @hzh0425, 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 type inconsistency within the SWARadixCache that was causing CUDA device-side assertions and invalid memory access. By explicitly casting KV indices to torch.int64 before they are stored in the radix tree, the change ensures data type alignment across the system, thereby enhancing the stability and correctness of the caching mechanism.

Highlights

• Problem Identification: A CUDA device-side assertion was occurring due to a type mismatch where KV indices were stored as int32 in the radix tree but expected as int64 in Triton kernel operations, leading to invalid memory access.
• Solution Implemented: The page_aligned_kv_indices are now explicitly converted to torch.int64 using to(dtype=torch.int64, copy=True) before being stored in the radix tree, ensuring type consistency.
• Affected Functions: This type conversion has been applied in both the cache_finished_req and cache_unfinished_req methods within the SWARadixCache class.

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 fixes a type inconsistency issue for KV indices in SWARadixCache by explicitly converting them to torch.int64. This prevents a potential CUDA assertion failure and ensures the stability of the data pipeline.

I've added a couple of suggestions to refactor the modified code blocks. These changes will reduce code duplication and improve readability and maintainability without altering the logic.