GRM Llama3.2 3B Rewardmodel Ft

🚀 GRM-Llama3.2-3B奖励模型

本奖励模型是基于Transformer架构的奖励模型，通过对基础模型进行微调，在奖励模型基准测试中取得了优异的成绩，为大语言模型的奖励评估提供了有效的解决方案。

🚀 快速开始

本奖励模型在 reward-bench 上取得了90.9的分数。它是基于 Ray2333/GRM-llama3.2-3B-sftreg 模型，使用经过净化处理的 Skywork偏好数据集v0.2 进行微调得到的。我们得到了一个SOTA 3B奖励模型，它可以超越一系列8B奖励模型，甚至在作为评判器时超过gpt4/gemini。

你可以在🤗 hugging face 查看我们的GRM系列模型，在 Arxiv 查看我们的论文，在 Github 查看我们的代码仓库。

✨ 主要特性

• 高性能：在奖励模型基准测试中，该模型在小于7B的模型中表现出色。
• 小模型大能量：3B参数的模型能够超越一系列8B奖励模型，甚至在评判任务中超过gpt4/gemini。

📦 安装指南

此部分原文档未提及具体安装步骤，故跳过。

💻 使用示例

基础用法

import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

device = 'cuda:0'
# load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained('Ray2333/GRM-Llama3.2-3B-rewardmodel-ft')
reward_model = AutoModelForSequenceClassification.from_pretrained(
                'Ray2333/GRM-Llama3.2-3B-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)

# The encode_plus may add another bos token though no impact on the final performance, but you can also avoid this by using the following code:
# tokens =  tokenizer.apply_chat_template(message, tokenize=True, return_dict=True, **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()

📚 详细文档

评估

我们在 奖励模型基准测试 上对GRM-Llama3.2-3B-rewardmodel-ft进行了评估，该模型在小于7B的模型中表现强劲。

⚠️ 重要提示

使用reward bench进行评估时，请添加 '--not_quantized' 以避免性能下降。

引用

如果你发现这个模型对你的研究有帮助，请引用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}
}

📄 许可证

本项目采用Apache 2.0许可证。