库名称: transformers 模型名称: outputs/ablation-141-a128.dpo.armorm.rp-shisa-v2-llama-3.1-8b 标签:

• 训练生成 许可证: license

模型卡片: outputs/ablation-141-a128.dpo.armorm.rp-shisa-v2-llama-3.1-8b

该模型是基于None进行微调的版本。 它使用TRL框架进行训练。

快速开始

from transformers import pipeline

question = "如果你有一台时光机，但只能选择回到过去或前往未来一次且无法返回，你会如何选择？为什么？"
generator = pipeline("text-generation", model="None", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])

训练过程

该模型采用DPO方法训练，该方法源自论文《直接偏好优化：你的语言模型其实是奖励模型》。

框架版本

• TRL: 0.15.1
• Transformers: 4.50.0
• Pytorch: 2.6.0
• Datasets: 3.4.1
• Tokenizers: 0.21.1

引用文献

引用DPO方法:

@inproceedings{rafailov2023direct,
    title        = {{直接偏好优化：你的语言模型其实是奖励模型}},
    author       = {Rafael Rafailov and Archit Sharma and Eric Mitchell and Christopher D. Manning and Stefano Ermon and Chelsea Finn},
    year         = 2023,
    booktitle    = {第36届神经信息处理系统年会论文集：NeurIPS 2023, 2023年12月10-16日, 美国路易斯安那州新奥尔良},
    url          = {http://papers.nips.cc/paper_files/paper/2023/hash/a85b405ed65c6477a4fe8302b5e06ce7-Abstract-Conference.html},
    editor       = {Alice Oh and Tristan Naumann and Amir Globerson and Kate Saenko and Moritz Hardt and Sergey Levine},
}

引用TRL框架:

@misc{vonwerra2022trl,
	title        = {{TRL：基于Transformer的强化学习}},
	author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
	year         = 2020,
	journal      = {GitHub仓库},
	publisher    = {GitHub},
	howpublished = {\url{https://github.com/huggingface/trl}}
}