Fix: mrope_position_delta missing in PD disaggregation decode mode for qwen-vl series

[LeoGoTi]

This PR has been closed.
Check for this PR as it's handling this issue.

Motivation

This PR aims to fix a bug where mrope_position_delta is missing on PD-disaggregation setup, for decode mode only.

In brief:

• Qwen-vl series models implement the mrope algorithm stage one for mrope_positions and mrope_position_delta.
• mrope_position_delta is necessary for generating subsequent positions during the decode stage, and is not accessible when PD disaggregation is deployed.

Modifications

• Add fields named decode_mrope_position_delta to the Req and ModelWorkerBatch dataclasses.
• On _compute_mrope_positions calculation, if the delta values are available from the decode_mrope_position_delta field, we use it directly.
• Add tensor allocation named mrope_position_delta in MetaDataBuffers, serving as a kv-meta part being transferred from prefill to decode. This is only valid when we register MetaDataBuffers using is_mrope_enabled=True.

Checklist

• Format your code according to the Code Formatting with Pre-Commit.
• Add unit tests as outlined in the Running Unit Tests.
• Update documentation / docstrings / example tutorials as needed, according to Writing Documentation.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed, according to Benchmark and Profiling and Accuracy Results.
• For reviewers: If you haven't made any contributions to this PR and are only assisting with merging the main branch, please remove yourself as a co-author when merging the PR.
• Please feel free to join our Slack channel at https://slack.sglang.ai to discuss your PR.

[gemini-code-assist]

Summary of Changes

Hello @LeoGoTi, 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 critical bug affecting Qwen-VL series models when operating in Paged-Disaggregation (PD) decode mode. The core issue was the unavailability of mrope_position_delta, a value essential for correctly computing subsequent token positions during the decode phase. The changes introduce the necessary plumbing to ensure this delta value is properly transferred from the prefill stage to the decode stage via the MetaDataBuffers and made accessible throughout the request and batch processing pipeline.

Highlights

• Data Structure Enhancement: Introduced decode_mrope_position_delta field to Req, ModelWorkerBatch, and ForwardBatch dataclasses to carry the necessary position delta information for Qwen-VL models in decode mode.
• Metadata Buffer Integration: Modified the MetaDataBuffers class to conditionally allocate, store, and retrieve the mrope_position_delta tensor, enabling its transfer from prefill to decode stages when mROPE is enabled.
• Positional Calculation Update: Updated the _compute_mrope_positions function to utilize the newly available decode_mrope_position_delta for accurate positional calculations during the decode phase, specifically for Paged-Disaggregation (PD) setups, and added a fallback for warmup scenarios.

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 is currently in preview and 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 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 fixes an issue where mrope_position_delta is missing in PD disaggregation decode mode for qwen-vl series models. The changes involve adding fields to dataclasses and using the delta values directly when available. A critical bug was identified in the m-rope position calculation, along with maintainability issues.

[LeoGoTi]

Closed this PR as another active PR is doing this more elegantly.