refactor: update trtllm-gen attn interface

[yyihuang]

📌 Description

Calculate mMaxSeqLenKv in kernel launch as page_size * num_pages.

🔍 Related Issues

🚀 Pull Request Checklist

Thank you for contributing to FlashInfer! Before we review your pull request, please make sure the following items are complete.

✅ Pre-commit Checks

• I have installed pre-commit by running pip install pre-commit (or used your preferred method).
• I have installed the hooks with pre-commit install.
• I have run the hooks manually with pre-commit run --all-files and fixed any reported issues.

If you are unsure about how to set up pre-commit, see the pre-commit documentation.

🧪 Tests

• Tests have been added or updated as needed.
• All tests are passing (unittest, etc.).

Reviewer Notes

[gemini-code-assist]

Summary of Changes

Hello @yyihuang, 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 refactors the interface of TensorRT-LLM batch attention functions by removing redundant sequence length parameters. These values are now derived internally from the KV cache and block table shapes, leading to cleaner and more robust function signatures across the decode and prefill operations.

Highlights

• Interface Refinement: I've removed the max_seq_len and max_kv_len parameters from the trtllm_batch_decode_with_kv_cache and trtllm_batch_context_with_kv_cache functions, respectively. These parameters were redundant as their values can be inferred.
• Dynamic Parameter Derivation: The max_seq_len and max_kv_len values are now dynamically computed within the respective functions using the k_cache.shape[3] (page_size) and block_tables.shape[1] (num_pages), simplifying the function signatures.
• Test Updates: I've updated the corresponding test calls in tests/test_trtllm_gen_context.py and tests/test_trtllm_gen_decode.py to align with the modified function signatures, removing the now-unnecessary max_seq_len and max_kv_len arguments.

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 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 or fill out our survey 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 refactors the trtllm-gen attention interfaces to derive max_seq_len and max_kv_len from the shapes of the cache and block table tensors, rather than passing them as arguments. This is a positive change that improves the API by reducing redundant parameters.

However, I've identified a critical issue in how page_size is calculated in both flashinfer/decode.py and flashinfer/prefill.py. An incorrect tensor dimension is used, which will lead to incorrect calculations for max_seq_len and max_kv_len, likely causing runtime errors or incorrect attention results. I've provided suggestions to fix this indexing error.

Once these critical issues are addressed, the refactoring will be a solid improvement to the codebase.

[gemini-code-assist]

There appears to be an indexing error in the calculation of page_size. Based on the function's docstring and the tensor unpacking logic, k_cache has a shape of [num_pages, num_kv_heads, page_size, head_dim] when kv_layout is HND.

The current implementation uses k_cache.shape[3], which corresponds to head_dim, not page_size. The correct index for page_size is 2.

This will cause an incorrect max_seq_len to be calculated, which is a critical bug that could lead to incorrect kernel behavior or memory access errors.

Suggested change

	page_size = k_cache.shape[3]
	page_size = k_cache.shape[2]

[gemini-code-assist]

There is an indexing error when calculating page_size. According to the function's docstring and tensor unpacking logic, k_cache has a shape of [num_pages, num_kv_heads, page_size, head_dim] when kv_layout is HND. Therefore, page_size should be accessed via k_cache.shape[2].

The current code uses k_cache.shape[3], which incorrectly retrieves the head_dim. This will lead to an incorrect max_kv_len and is a critical bug.

Suggested change

	page_size = k_cache.shape[3]
	page_size = k_cache.shape[2]