[Kwai-Keye]

Purpose

• Support Keye-VL-8B-Preview

Example Serving Command

• The model can be served with

vllm serve \
    Kwai-Keye/Keye-VL-8B-Preview \
    --tensor-parallel-size 8 \
    --enable-prefix-caching \
    --gpu-memory-utilization 0.8 \
    --host 0.0.0.0 \
    --port 8001 \
    --trust-remote-code

Thank you for your time on reviewing this PR :)

[github-actions]

👋 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 fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

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 ready label to the PR or enable auto-merge.

🚀

[gemini-code-assist]

Summary of Changes

Hello @Kwai-Keye, 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 introduces full support for the Keye-VL-8B-Preview model within vLLM. It includes the necessary components for processing multimodal inputs (images and videos), integrating a vision encoder with the language model, and registering the new model for seamless deployment and serving.

Highlights

• New Model Support: Adds comprehensive support for the Keye-VL-8B-Preview Vision-Language Model (VLM), enabling it to be served by vLLM.
• Multimodal Processing: Implements a dedicated KeyeProcessor and related multimodal data handling components to correctly process image and video inputs for the Keye model, including tokenization and embedding integration.
• Vision Backbone Integration: Integrates a Siglip-based vision encoder and a projector module to extract and transform visual features, which are then merged with the language model's embeddings.
• Model Registration: Registers the KeyeForConditionalGeneration model within vLLM's internal registry, making it discoverable and loadable via its model name.
• Token Handling: Updates chat_utils.py to ensure the Keye model correctly uses Qwen-style vision tokens (<|vision_start|><|image_pad|><|vision_end|>, etc.) for multimodal inputs.

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

The pull request introduces support for the Keye-VL-8B-Preview model, including its vision and multimodal processing components. The changes involve adding new model definitions and integrating them into the existing vLLM framework. The review identifies several areas for improvement related to code clarity, consistency, error handling, and adherence to best practices.

[DarkLight1337]

Thanks for adding this model. Left a few initial comments

[Isotr0py]

LGTM now! Would be good if we can add SDPA support for ViT attention in a following PR.

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @Kwai-Keye.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[Kwai-Keye]

LGTM now! Would be good if we can add SDPA support for ViT attention in a following PR.

Considering that xformers is a default dependency listed in requirements/cuda.txt, we've prioritized supporting the xformers backend first~ Kwai-Keye@475d004