[model-gateway] multimodality initialization

[slin1237]

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.

[gemini-code-assist]

Summary of Changes

Hello @slin1237, 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 lays the groundwork for supporting multimodal models within the sgl-router. It introduces the necessary infrastructure to process and track various media types, particularly images, alongside text inputs. The changes enable the router to interact with Python-based HuggingFace processors for multimodal model inference, ensuring compatibility and extensibility for future multimodal model integrations.

Highlights

• Multimodal Feature Integration: Introduced a new python-mm Cargo feature to enable multimodal capabilities within the sgl-router.
• New Dependencies: Added several new Rust dependencies including pyo3 for Python interoperability, image for image processing, base64 for encoding/decoding, and serde_bytes for efficient byte serialization. Corresponding Python dependencies like pillow, transformers, torch, and numpy were also added.
• Multimodal Core Modules: Implemented new Rust modules under sgl-router/src/multimodal to handle various aspects of multimodal processing, including error handling, media fetching, model registry, and input tracking.
• Python Bridge for Model Processing: Created a Python shim (mm_processor_shim.py) and a Rust-Python bridge (python_bridge.rs) to leverage HuggingFace processors for multimodal input processing, mirroring vLLM's invocation.
• Model-Specific Processing Logic: Developed a ModelRegistry and ModelProcessorSpec trait to define and manage model-specific logic for multimodal inputs, with initial support for Llava, Qwen-VL, and Phi-3-Vision models.
• Multimodal Input Tracking: Implemented an AsyncMultiModalTracker to manage and process diverse multimodal content parts (text, image URLs, image data) within a conversation, including placeholder insertion and media fetching.

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 significant new functionality for multimodal support, which is a substantial and well-architected addition. The new modules for media handling, model-specific processing, and input tracking are logically structured. The use of a Python bridge to interface with HuggingFace processors is a practical choice, and the inclusion of parity tests against vLLM is commendable for ensuring compatibility. My review identifies a few areas for improvement, including a potential performance issue with blocking I/O, the use of deprecated APIs, and a type inconsistency, which I've detailed in the comments.

[gemini-code-assist]

The use of std::fs::canonicalize within the new function is a blocking I/O operation. When this constructor is called from an asynchronous context, it can block the Tokio runtime's worker thread, leading to reduced throughput and potential performance bottlenecks. It is highly recommended to use the asynchronous equivalent, tokio::fs::canonicalize, and make the new function async. This will ensure that the application remains fully non-blocking. Note that this change will require updating the call sites of MediaConnector::new to await its result.

pub async fn new(client: Client, config: MediaConnectorConfig) -> Result<Self, MediaConnectorError> {
    let allowed_domains = config.allowed_domains.map(|domains| {
        domains
            .into_iter()
            .map(|d| d.to_ascii_lowercase())
            .collect::<HashSet<_>>()
    });

    let allowed_local_media_path = if let Some(path) = config.allowed_local_media_path {
        Some(tokio::fs::canonicalize(path).await?)
    } else {
        None
    };

    Ok(Self {
        client,
        allowed_domains,
        allowed_local_media_path,
        fetch_timeout: config.fetch_timeout,
    })
}