Fix missing dp_max_padding argument in set_dp_buffer_len

[Chen-0210]

Motivation

PR #12572 added a new argument dp_max_padding to set_dp_buffer_len, which caused #12796 .
Fixes #12796.

Modifications

Pass the missing dp_max_padding argument to set_dp_buffer_len.

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 @Chen-0210, 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 resolves a critical bug stemming from a recent update that introduced a new argument, dp_max_padding, to the set_dp_buffer_len function. By correctly supplying this previously missing argument, the change ensures the proper operation of the deep attention padding mechanism, preventing potential issues and maintaining system stability.

Highlights

• Bug Fix: Addresses a bug where the set_dp_buffer_len function was missing the dp_max_padding argument, which was introduced in a previous pull request (Register allgather/reducescatter buffers with symm memory #12572) and led to issue [Bug] Piecewise Cuda Graph Issue in v0.5.5 #12796.
• Argument Passing: The dp_max_padding argument is now correctly passed to set_dp_buffer_len using forward_batch.dp_padding_mode.is_max_len() within the capture_one_batch_size method.

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 crash by passing the missing dp_max_padding argument to set_dp_buffer_len. However, I've identified a related latent bug where global_dp_buffer_len is None, which would cause a TypeError when data parallelism is enabled. My suggested change addresses both the original issue and this latent bug by correctly calculating global_dp_buffer_len and also passing the global_num_tokens argument, which is necessary for data parallel attention.

[gemini-code-assist]

While adding the dp_max_padding argument fixes the immediate TypeError, there's a latent bug here. The global_dp_buffer_len variable is initialized to None at line 331. When data parallelism is used (dp_size > 1), this None value will be passed to set_dp_buffer_len and eventually cause a TypeError inside get_global_dp_buffer when it's used to allocate a tensor.

To make this robust, we should calculate the correct global_dp_buffer_len here, which should be num_tokens * self.dp_size. Additionally, the global_num_tokens argument should be passed for data parallel attention to work correctly during graph capture.

Suggested change

	set_dp_buffer_len(global_dp_buffer_len, num_tokens)
	set_dp_buffer_len(
	num_tokens * self.dp_size,
	num_tokens,
	forward_batch.dp_padding_mode.is_max_len(),
	[num_tokens] * self.dp_size,
	)