QiWang98

/

VideoRFT-3B

like 0

This PR significantly enhances the model card for VideoRFT, aligning it with Hugging Face best practices.

Key changes include:

• Populated Content: Added a comprehensive description, abstract, methodology details, dataset information, setup instructions, training guidance, evaluation procedure, and a ready-to-use inference code snippet, all extracted from the paper and GitHub repository.
• Metadata Updates:\n - Changed pipeline_tag from visual-question-answering to video-text-to-text to better reflect the model's general video reasoning capabilities.
  • Added library_name: transformers as the model is compatible with the 🤗 Transformers library, enabling the automated "Use in Transformers" widget.
• Links: Ensured proper linking to the Hugging Face paper page (VideoRFT: Incentivizing Video Reasoning Capability in MLLMs via Reinforced Fine-Tuning) and the GitHub repository (https://github.com/QiWang98/VideoRFT).

These updates will make the model more discoverable and easier to use for the community.