CodeGoat24

/

UnifiedReward-qwen-7b

like 5

This PR improves the model card for UnifiedReward-qwen-7b, clarifying its role as a reward model in the context of the Pref-GRPO: Pairwise Preference Reward-based GRPO for Stable Text-to-Image Reinforcement Learning paper.

Specifically, it addresses the following points:

• Corrected Paper Link: Updated the paper link in the content to https://huggingface.co/papers/2508.20751 as per the prompt's instruction, which describes the Pref-GRPO framework in which this reward model is used.
• Updated Project Page Link: Pointed the project page to the more specific https://codegoat24.github.io/UnifiedReward/Pref-GRPO.
• Added GitHub Link: Included a direct link to the Pref-GRPO code at https://github.com/CodeGoat24/Pref-GRPO.
• Added Pipeline Tag: Set pipeline_tag: image-text-to-text to accurately reflect the model's input (image, text) and output (text evaluation). This helps users discover the model more easily.
• Added Library Name: Specified library_name: transformers as the model uses the Hugging Face Transformers library for inference, enabling the "Use in Transformers" widget.
• Made Sample Usage Runnable: Added import requests to the Python snippet to ensure the example code is runnable out-of-the-box.
• Updated Citation: Replaced the outdated BibTeX entry with the correct one for the Pref-GRPO paper.
• Updated Quick Start GitHub Link: Corrected the GitHub link in the Quick Start section to point to the Pref-GRPO repository.

These changes enhance the discoverability, clarity, and completeness of the model card.