README.md

---
base_model:
- Qwen/Qwen2.5-VL-7B-Instruct
datasets:
- CodeGoat24/HPD
- CodeGoat24/LiFT-HRA
- CodeGoat24/OIP
- CodeGoat24/EvalMuse
- CodeGoat24/ShareGPTVideo-DPO
- CodeGoat24/LLaVA-Critic-113k
- CodeGoat24/VideoDPO
license: mit
pipeline_tag: image-text-to-text
library_name: transformers
---

# UnifiedReward-qwen-7B: A Reward Model for Pref-GRPO

We are actively gathering feedback from the community to improve our models. **We welcome your input and encourage you to stay updated through our repository**!!

## Model Summary

`UnifiedReward-qwen-7b` is the first unified reward model based on [Qwen/Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct) for multimodal understanding and generation assessment. It enables both pairwise ranking and pointwise scoring, and is notably employed for vision model preference alignment within the **Pref-GRPO** framework.

This model is a key component of the research presented in the paper [**Pref-GRPO: Pairwise Preference Reward-based GRPO for Stable Text-to-Image Reinforcement Learning**](https://huggingface.co/papers/2508.20751).

For further details, please refer to the following resources:
- 📰 Paper: [Pref-GRPO: Pairwise Preference Reward-based GRPO for Stable Text-to-Image Reinforcement Learning](https://huggingface.co/papers/2508.20751)
- 🪐 Project Page: https://codegoat24.github.io/UnifiedReward/Pref-GRPO
- 💻 Code: https://github.com/CodeGoat24/Pref-GRPO
- 🤗 Model Collections: https://huggingface.co/collections/CodeGoat24/unifiedreward-models-67c3008148c3a380d15ac63a
- 🤗 Dataset Collections: https://huggingface.co/collections/CodeGoat24/unifiedreward-training-data-67c300d4fd5eff00fa7f1ede
- 👋 Point of Contact: [Yibin Wang](https://codegoat24.github.io)


## 🏁 Compared with Current Reward Models

| Reward Model | Method| Image Generation | Image Understanding | Video Generation | Video Understanding
| :-----: | :-----: |:-----: |:-----: | :-----: | :-----: |
| [PickScore](https://github.com/yuvalkirstain/PickScore) |Point | √ | | ||
| [HPS](https://github.com/tgxs002/HPSv2) | Point | √ | |||
| [ImageReward](https://github.com/THUDM/ImageReward) | Point| √| |||
| [LLaVA-Critic](https://huggingface.co/lmms-lab/llava-critic-7b) | Pair/Point | | √ |||
| [IXC-2.5-Reward](https://github.com/InternLM/InternLM-XComposer) | Pair/Point | | √ ||√|
| [VideoScore](https://github.com/TIGER-AI-Lab/VideoScore) | Point | | |\u221a ||
| [LiFT](https://github.com/CodeGoat24/LiFT) | Point | | |\u221a| |
| [VisionReward](https://github.com/THUDM/VisionReward) | Point |√ | |\u221a||
| [VideoReward](https://github.com/KwaiVGI/VideoAlign) | Point | | |\u221a ||
| UnifiedReward (Ours) | Pair/Point | √ | √ |\u221a|\u221a|


### Quick Start
All pair rank and point score inference codes are provided in our [GitHub repository](https://github.com/CodeGoat24/Pref-GRPO).

We take image understanding assessment as example here:
~~~python
import json
import random
import torch
import tqdm
from PIL import Image
import warnings
import os
import requests # Added for image download in example
from transformers import AutoProcessor, AutoTokenizer, Qwen2_5_VLForConditionalGeneration
from qwen_vl_utils import process_vision_info

warnings.filterwarnings("ignore")

model_path = "CodeGoat24/UnifiedReward-qwen-7b"
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    model_path, torch_dtype="auto", device_map="auto"
)
processor = AutoProcessor.from_pretrained(model_path)


url = "https://github.com/LLaVA-VL/blog/blob/main/2024-10-03-llava-critic/static/images/critic_img_seven.png?raw=True"
image = Image.open(requests.get(url, stream=True).raw)

prompt_text = f'Given an image and a corresponding question, please serve as an unbiased and fair judge to evaluate the quality of the answers provided by a Large Multimodal Model (LMM). Determine which answer is better and explain your reasoning with specific details. Your task is provided as follows:\
Question: [What this image presents?]\
The first response: [The image is a black and white sketch of a line that appears to be in the shape of a cross. The line is a simple and straightforward representation of the cross shape, with two straight lines intersecting at a point.]\
The second response: [This is a handwritten number seven.]\
ASSISTANT:\
'

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": image},
            {"type": "text", "text": prompt_text},
        ],
    }
]

chat_input = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
image_inputs, video_inputs = process_vision_info(messages)

inputs = processor(
    text=[chat_input],
    images=image_inputs,
    videos=video_inputs,
    return_tensors="pt",
    padding=True
).to("cuda")

with torch.no_grad():
    generated_ids = model.generate(**inputs, max_new_tokens=4096)
generated_trimmed = [
    out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output = processor.batch_decode(generated_trimmed, skip_special_tokens=True)[0]


print(output)
~~~


## Citation

```bibtex
@article{Pref-GRPO&UniGenBench,
  title={Pref-GRPO: Pairwise Preference Reward-based GRPO for Stable Text-to-Image Reinforcement Learning},
  author={Wang, Yibin and Li, Zhimin and Zang, Yuhang and Zhou, Yujie and Bu, Jiazi and Wang, Chunyu and Lu, Qinglin, and Jin, Cheng and Wang, Jiaqi},
  journal={arXiv preprint arXiv:2508.20751},
  year={2025}
}
```