HuRef:面向大语言模型的人类可读指纹
HuRef: HUman-REadable Fingerprint for Large Language Models
AI总结:
针对大语言模型基础模型溯源与版权保护问题,提出 HuRef 指纹,利用 Transformer 参数不变量、StyleGAN2 图像编码和 ZKP 实现黑盒下可验证且不泄露参数的唯一识别。
摘要:
由于大语言模型(LLM)的训练需要消耗大量资源,并附带经过精心设计的许可证,保护其版权已变得至关重要。然而,由于参数可能被改动,识别一个 LLM 的原始基础模型颇具挑战。在本研究中,我们提出 HuRef,这是一种面向 LLM 的人类可读指纹,可在不干扰训练且不向公众暴露模型参数的情况下唯一识别基础模型。我们首先观察到,在预训练期间模型收敛之后,LLM 参数的向量方向保持稳定,并且在后续训练步骤(包括继续预训练、监督微调和 RLHF)中只会受到可忽略的扰动,这使其成为识别基础模型的充分条件。我们通过使用一个额外项继续训练 LLM、驱使模型参数方向偏离并导致模型受损,验证了该条件的必要性。然而,这一方向容易受到维度置换或矩阵旋转等简单攻击的影响,这些攻击会在不影响性能的情况下显著改变方向。为解决该问题,我们利用 Transformer 结构,系统分析潜在攻击,并定义三个用于识别 LLM 基础模型的不变量。由于存在信息泄露的潜在风险,我们不能直接发布这些不变量。相反,我们使用编码器将其映射为一个高斯向量,再利用 StyleGAN2 将其转换为自然图像,最后发布该图像。在我们的黑盒设置中,所有指纹生成步骤均由 LLM 所有者在内部完成。为确保发布的指纹经过诚实生成,我们引入了零知识证明(ZKP)。跨多种 LLM 的实验结果证明了本方法的有效性。代码可在 https://github.com/LUMIA-Group/HuRef 获取。
英文摘要:
Protecting the copyright of large language models (LLMs) has become crucial due to their resource-intensive training and accompanying carefully designed licenses. However, identifying the original base model of an LLM is challenging due to potential parameter alterations. In this study, we introduce HuRef, a human-readable fingerprint for LLMs that uniquely identifies the base model without interfering with training or exposing model parameters to the public. We first observe that the vector direction of LLM parameters remains stable after the model has converged during pretraining, with negligible perturbations through subsequent training steps, including continued pretraining, supervised fine-tuning, and RLHF, which makes it a sufficient condition to identify the base model. The necessity is validated by continuing to train an LLM with an extra term to drive away the model parameters' direction and the model becomes damaged. However, this direction is vulnerable to simple attacks like dimension permutation or matrix rotation, which significantly change it without affecting performance. To address this, leveraging the Transformer structure, we systematically analyze potential attacks and define three invariant terms that identify an LLM's base model. Due to the potential risk of information leakage, we cannot publish invariant terms directly. Instead, we map them to a Gaussian vector using an encoder, then convert it into a natural image using StyleGAN2, and finally publish the image. In our black-box setting, all fingerprinting steps are internally conducted by the LLMs owners. To ensure the published fingerprints are honestly generated, we introduced Zero-Knowledge Proof (ZKP). Experimental results across various LLMs demonstrate the effectiveness of our method. The code is available at https://github.com/LUMIA-Group/HuRef.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。