[bugfix] fix megatron non-padding_free qwen3_vl cp

[Jintao-Huang]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello @Jintao-Huang, 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 bug within the Megatron framework, specifically targeting issues that arise when handling 'non-padding_free' Qwen3_VL models in a context parallel setup. The changes ensure that the system gracefully manages situations where sequence packing parameters are absent, preventing runtime errors and enhancing the stability of batch processing under these specific parallelization configurations.

Highlights

• Robustness for packed_seq_params: Modified calls to split_cp_inputs across two files to safely handle cases where packed_seq_params might be None, preventing potential AttributeErrors during execution.
• Type Hint Update: The split_cp_inputs function signature was updated to explicitly allow cu_seqlens to be Optional[torch.Tensor], improving code clarity and type safety for scenarios where sequence lengths might not be provided.
• Context Parallelism Batch Handling: Adjusted the get_batch_on_this_cp_rank function to continue processing batches even when packed_seq_params is None, ensuring correct behavior in 'non-padding_free' scenarios for context parallelism.

🧠 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 aims to fix a bug related to context parallelism for non-padding-free scenarios in Megatron models. The changes correctly identify that packed_seq_params can be None and update the call sites to pass None for cu_seqlens. However, the implementation of split_cp_inputs is not updated to handle this None case, which would lead to a runtime error. I have provided a critical review comment with a suggested fix that addresses this issue, refactors the function for better maintainability, and adds a validation check.

[Jintao-Huang]

/gemini review

[gemini-code-assist]

Code Review

This pull request addresses a bug related to context parallelism (CP) for non-padding-free sequences in Megatron, specifically for the qwen3_vl model. The changes are well-implemented and resolve the issue effectively. The core of the fix involves modifying split_cp_inputs to handle both padding-free (packed) and non-padding-free sequences, which simplifies the logic in several places. The related adjustments in qwen3_vl.py, mm_gpt_model.py, and init.py are consistent and correct. Overall, this is a solid bugfix that improves the robustness of the codebase. I have a few minor suggestions to enhance code readability by using getattr for safer attribute access.

[Jintao-Huang]

/gemini review

[gemini-code-assist]

Code Review

This pull request refactors the handling of context parallelism and packed sequences, primarily by making the cu_seqlens_q parameter optional in various split_cp_inputs calls across qwen3_vl.py, mm_gpt_model.py, and utils.py through the use of getattr(packed_seq_params, 'cu_seqlens_q', None). The split_cp_inputs function itself in utils.py was updated to gracefully handle None values for cu_seqlens, treating the input as a single sequence. Additionally, the direct slicing logic for rotary position embeddings based on context parallel rank was removed from _patch_mrope in init.py, along with related imports, and the mcore_get_batch_on_this_cp_rank utility and its usage were removed from utils.py. An assertion for packed_seq_params being non-None was also removed in qwen3_vl.py.