Kwai-Keye

/

Keye-VL-8B-Preview

like 80

This PR significantly enhances the model card for Kwai Keye-VL by:

• Updating the pipeline_tag: Changed from image-text-to-text to video-text-to-text. This change accurately reflects the model's primary strength and focus on video understanding, as highlighted in the paper's abstract and experimental results. This will improve its discoverability for users searching for video-capable models on the Hub.
• Adding the paper abstract: The full abstract from the technical report has been added to provide a comprehensive overview of the model's capabilities, development, and innovative training recipe directly on the model page.
• Including a direct code link: A link to the official GitHub repository (https://github.com/Kwai-Keye/Keye) has been added to the top navigation bar for easier access to the model's codebase and utilities.
• Updating the citation: The BibTeX entry has been updated to the more complete version found in the official GitHub repository, ensuring proper attribution to the technical report.

These changes make the model card more informative, accurate, and user-friendly.