GRM Gemma2 2B Rewardmodel Ft

🚀 GRM-Gemma2-2B Reward Model

This project presents a high - performance reward model that addresses the challenges in reward evaluation for large - language models. It offers a lightweight yet powerful solution that outperforms many larger models in the reward - benchmarking tasks.

🚀 Quick Start

This reward model achieves a score of 88.4 on reward - bench. It is finetuned from the [Ray2333/GRM - Gemma2 - 2B - sftreg](https://huggingface.co/Ray2333/GRM - Gemma2 - 2B - sftreg) using the decontaminated [Skywork preference dataset v0.2](https://huggingface.co/datasets/Skywork/Skywork - Reward - Preference - 80K - v0.2). We've obtained a SOTA 2B reward model that can outperform a series of 8B reward models and even surpass gpt4/gemini as a judge.

Check our GRM series at 👋[hugging face](https://huggingface.co/collections/Ray2333/grm - 66882bdf7152951779506c7b), our paper at Arxiv, and github repo at [Github](https://github.com/YangRui2015/Generalizable - Reward - Model).

✨ Features

• High Performance: Achieves excellent scores on the reward - bench, outperforming many larger models.
• Lightweight: Based on a 2B model, offering efficiency without sacrificing performance.
• Generalizable: Can be used as a judge to evaluate various responses effectively.

📦 Installation

No specific installation steps are provided in the original document.

💻 Usage Examples

Basic Usage

import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

device = 'cuda:0'
# load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained('Ray2333/GRM-Gemma2-2B-rewardmodel-ft')
reward_model = AutoModelForSequenceClassification.from_pretrained(
                'Ray2333/GRM-Gemma2-2B-rewardmodel-ft', torch_dtype=torch.float16, 
                device_map=device,
                )
message = [
  {'role': 'user', 'content': "I'm going to go out to a movie, but I need someone to chat with my daughter and pretend to be me while she's home alone.  But I can't do that while I'm at the movie.  Can you help by impersonating me by chat with her?"},
  {'role': 'assistant', 'content': "Sorry, I'm not comfortable impersonating you in that way.  I'm not willing to behave so dishonestly.  Maybe you can just find a way to bring her to the movie, or you can find a babysitter?"}
]
message_template = tokenizer.apply_chat_template(message, tokenize=False)
# it will look like this: "<bos><start_of_turn>user\nI'm going to go out to a movie, but I need someone to chat with my daughter and pretend to be me while she's home alone.  But I can't do that while I'm at the movie.  Can you help by impersonating me by chat with her?<end_of_turn>\n<start_of_turn>model\nSorry, I'm not comfortable impersonating you in that way.  I'm not willing to behave so dishonestly.  Maybe you can just find a way to bring her to the movie, or you can find a babysitter?<end_of_turn>\n".

kwargs = {"padding": 'longest', "truncation": True, "return_tensors": "pt"}
tokens = tokenizer.encode_plus(message_template, **kwargs)

with torch.no_grad():
  reward_tensor = reward_model(tokens["input_ids"][0].view(1,-1).to(device), attention_mask=tokens["attention_mask"][0].view(1,-1).to(device))[0]
  reward = reward_tensor.cpu().detach().item()

Advanced Usage

There is no advanced usage example provided in the original document.

📚 Documentation

Evaluation

We evaluate GRM - Gemma2 - 2B - rewardmodel - ft on the [reward model benchmark](https://huggingface.co/spaces/allenai/reward - bench), where it achieved SOTA performance among models smaller than 3B.

When evaluated using reward bench, please add '--not_quantized' to avoid performance drop.

🔧 Technical Details

No technical details are provided in the original document.

📄 License

The model is released under the Apache - 2.0 license.

📖 Citation

If you find this model helpful for your research, please cite GRM

@inproceedings{yang2024regularizing,
  title={Regularizing Hidden States Enables Learning Generalizable Reward Model for LLMs},
  author={Yang, Rui and Ding, Ruomeng and Lin, Yong and Zhang, Huan and Zhang, Tong},
  booktitle={Advances in Neural Information Processing Systems},
  year={2024}
}