MTP Multimodal compatible

[jhvmhg]

PR type

• Bug Fix
• New Feature
• Document Updates
• More Models or Datasets Support

PR information

Write the detail information belongs to this PR.

Experiment results

Paste your experiment result here(if needed).

[gemini-code-assist]

Summary of Changes

Hello, 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 enhances the multimodal compatibility within the Megatron framework by introducing more flexible handling of rotary positional embeddings for packed sequences. It specifically addresses scenarios where position IDs might not conform to a single-element shape, which is crucial for supporting diverse input modalities. Additionally, a minor refactoring improves the readability of the recomputation logic.

Highlights

• Multimodal Compatibility: The strict assertion requiring position_ids.shape[0] to be 1 for packed sequences was removed, and new logic was added to handle cases where position_ids.shape[0] is not 1 by rolling the rotary_pos_emb.
• Code Readability: The condition for full recomputation during training was refactored into a dedicated boolean variable _do_recompute for improved clarity.
• TODO Removal: A TODO comment related to multimodal compatibility in the forward method was removed, indicating progress or completion of the related task.

🧠 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.

Changelog

• swift/megatron/init.py
  • Removed a TODO comment related to multimodal compatibility.
  • Updated the logic for applying rotary positional embeddings to packed sequences, allowing for position_ids with shapes other than [1, ...].
  • Refactored the recomputation condition into a new variable _do_recompute for better code clarity.

Activity

• No human activity has been recorded on this pull request yet.

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 makes Multi-Token Prediction (MTP) compatible with multimodal models. The core change in swift/megatron/init.py replaces a restrictive assertion on position_ids shape with more flexible logic. This new logic uses torch.roll to correctly align rotary position embeddings for next-token prediction when dealing with multimodal inputs in packed sequences. The changes appear to correctly implement the intended feature. I have added one suggestion to improve code clarity by adding a comment to an otherwise obscure part of the logic.

[Jintao-Huang]

Thank you for your suggestion. The changes from this PR have been integrated here:
#8313