[bugfix] fix qwen3_next packing(OOM); fix 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 critical issues related to memory management and parallelism in the Qwen3-Next model's Megatron Core implementation. It clarifies the implications of enabling sequence packing on padding-free operations, refines the handling of various parallelism groups (sequence, tensor, and context) to prevent OOM errors and ensure correct data distribution, and updates the project's dependency installation script to support newer Python versions and improve package management.

Highlights

• Packing Behavior Clarification: Updated comments in mcore.sh to clarify that enabling --packing in Qwen3-Next's mcore implementation disables the removal of padding_free during the linear-attention stage. This improves training efficiency and memory usage but allows different sequences to be visible to each other.
• Enhanced Parallelism Handling: Refined the logic for gather_from_sequence_parallel_region and scatter_to_sequence_parallel_region in qwen3_next.py to correctly determine and utilize parallelism groups based on sequence_parallel, tensor_model_parallel_size, and context_parallel_size, addressing potential OOM issues and ensuring proper data distribution.
• Padding-Free Unpadding Logic for Packed Inputs: Modified the padding_free unpadding mechanism in qwen3_next.py to explicitly bypass unpadding when args.packing is enabled. This change is crucial for maintaining efficiency and preventing unintended sequence interactions in packed inputs.
• Dependency Updates and Python Version Support: The install_all.sh script now supports Python 3.11 and adjusts the installation of transformers, trl, and peft by separating them from vllm for improved dependency management.

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 introduces fixes for the qwen3_next model, addressing an Out-Of-Memory (OOM) issue related to packing and a bug in context parallelism (cp). The changes in swift/megatron/model/gpt/qwen3_next.py correctly adjust the logic for sequence parallelism communication groups and conditionally skip sequence un-packing to prevent OOM. The dependency updates and comment clarifications in other files are also appropriate. My review includes one suggestion to improve the implementation's correctness and robustness.

[gemini-code-assist]

While the safe retrieval of attention_mask using kwargs.get() is a good improvement, you are accessing kwargs['attention_mask'] again inside the if block. It's better practice to use the attention_mask variable that you've already retrieved. This makes the code safer and more readable.

Suggested change

	attention_mask = kwargs['attention_mask'].sum(dim=(1, 3)) > 0
	attention_mask = attention_mask.sum(dim=(1, 3)) > 0