[Docs] Enhance SupportsMultiModal interface documentation #19701
Conversation
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add 🚀 |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Summary of Changes
Hello @russellb, 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 focuses entirely on improving the documentation for the SupportsMultiModal interface and its associated methods within vllm/model_executor/models/interfaces.py. The goal is to provide clearer guidance for developers implementing multimodal models in vLLM, detailing the purpose, implementation patterns, and system-level interactions of these critical components.
Highlights
- Enhanced
SupportsMultiModalInterface Documentation: I've added comprehensive class-level documentation to theSupportsMultiModalprotocol, explaining its purpose, integration with vLLM's V1 model runners, and recommended implementation patterns for multimodal models. supports_multimodalFlag Usage Clarification: The documentation for thesupports_multimodalflag has been enhanced to clearly outline its role in vLLM's inference engine, specifically for encoder cache allocation, multimodal-specific scheduling, LoRA compatibility warnings, and speculative decoding setup.- Detailed
get_multimodal_embeddings()Documentation: I've significantly improved the docstring forget_multimodal_embeddings(), detailing its core function in processing multimodal inputs, providing common implementation patterns, and specifying its usage contexts within V1 GPU/TPU model runners, encoder cache management, and profiling. - Expanded
get_language_model()Documentation: The documentation forget_language_model()now offers a more in-depth explanation of its purpose, common implementation examples across different model architectures (e.g.,self.language_model,self.model.decoder), and its critical uses in EAGLE speculative decoding, TPU compilation, and LoRA integration. - Comprehensive
get_input_embeddings()Documentation: I've updated theget_input_embeddings()docstring to thoroughly explain its role in converting input tokens to embeddings and merging them with multimodal embeddings. It now includes a standard implementation pattern, clarifies argument/return types, and highlights its usage in V1 model runners and TPU compilation, along with important notes on embedding ordering and the use ofmerge_multimodal_embeddings().
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
-
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. ↩
There was a problem hiding this comment.
Code Review
This pull request enhances the documentation for the SupportsMultiModal interface, detailing its purpose, usage, and implementation patterns for its methods. The changes improve clarity for developers. A suggestion regarding the return type hint for get_multimodal_embeddings has been noted for future consideration.
- Add comprehensive class-level documentation explaining interface purpose and V1 integration - Document supports_multimodal flag usage in inference engine (caching, scheduling, LoRA) - Enhance get_multimodal_embeddings() with implementation patterns and V1 usage contexts - Improve get_language_model() docs with architecture examples and V1 use cases - Expand get_input_embeddings() documentation with merge utility guidance and V1 integration Signed-off-by: Russell Bryant <rbryant@redhat.com>
russellb
force-pushed
the
supportsmultimodal-api-docs
branch
from
11ffae3 to
579eb2c
Compare
There was a problem hiding this comment.
Pull Request Overview
This PR enhances the SupportsMultiModal interface documentation by detailing its purpose, usage patterns, and integration points with vLLM V1.
- Expanded class-level doc to define workflow, usage, and implementation steps
- Documented
supports_multimodalflag usage in caching, scheduling, and LoRA - Enriched method-level docs (
get_multimodal_embeddings,get_language_model,get_input_embeddings) with usage contexts and example patterns
Comments suppressed due to low confidence (3)
vllm/model_executor/models/interfaces.py:49
- The
Implementation Patternsection is inline and lacks code block formatting. Consider using proper code block markup or Sphinx directives (e.g.,.. code-block:: python) to improve readability in generated documentation.
1. Inherit from SupportsMultiModal in your model class
vllm/model_executor/models/interfaces.py:81
- [nitpick] The use of
**kwargsforget_multimodal_embeddingsmakes the API ambiguous. Consider defining explicit parameters or a structured input type for supported modalities to enhance type safety and clarity.
Common Implementation Pattern:
vllm/model_executor/models/interfaces.py:62
- Newly documented behavior around LoRA compatibility warnings for multimodal models should be covered by tests. Add unit or integration tests to verify that warnings are emitted correctly when LoRA is applied.
- LoRA compatibility warnings
| Implementation Pattern: | ||
| 1. Inherit from SupportsMultiModal in your model class | ||
| 2. Implement all three required methods | ||
| 3. Use utilities like merge_multimodal_embeddings() for embedding |
There was a problem hiding this comment.
Can we use cross-references so they link to the corresponding location in API docs?
There was a problem hiding this comment.
Good suggestion. I'll look into how to do that properly.
Signed-off-by: Russell Bryant rbryant@redhat.com