zkFinGPT: Zero-Knowledge Proofs for Financial Generative Pre-trained Transformers

Ningjie Li, Keyi Wang, Xiaoli Zhi, Weiqin Tong, Xiao-Yang Liu

Keywords: FinGPT, Zero-knowledge proofs, use case, copyright, privacy

Abstract:

In recent years, large language models (LLMs) have been widely used and rapidly developed, with their performance increasing. However, due to the privacy of model parameters and input data, verifying the legitimacy of LLMs and the credibility of their outputs is a challenge. These issues are especially critical in the three financial use cases that we describe. In this paper, we propose zkFinGPT that introduces zero knowledge proofs to financial use cases. It enables both proof and verification while protecting data privacy. To be specific, we describe three financial use cases and how zkFinGPT can be used.
Experiments show that zkFinGPT has relatively low computational overhead, i.e., it generates a commitment file of $7.97$MB and takes $2.36$ seconds to verify the LLama-2-7B model.

Submission Number: 133