zkFinGPT: Zero-Knowledge Proofs for Financial Generative Pre-trained Transformers
Xiao-Yang Liu, Ningjie Li, Keyi Wang, Xiaoli Zhi, Weiqin Tong
TL;DR
zkFinGPT addresses the problem of verifying FinGPT outputs in high-stakes finance without exposing proprietary weights or training data. It leverages zero-knowledge proofs, specifically Multilinear Extensions and the Sumcheck Protocol, together with KZG polynomial commitments to prove $\bm{Y}=\bm{W}\bm{X}$ while keeping $\bm{W}$ and $\bm{X}$ private, with commitments stored on a blockchain for immutability. The approach is demonstrated across three financial use cases (copyright-corpus lawsuits, confidential exam-question testing, and trading-strategy privacy in contests) and evaluated for computational overhead, showing substantial prove-time costs and potential optimizations via quantization and GPU acceleration. The results indicate promising verifiability and privacy benefits but highlight practical adoption challenges due to current performance overheads, guiding future work toward more scalable implementations.
Abstract
Financial Generative Pre-trained Transformers (FinGPT) with multimodal capabilities are now being increasingly adopted in various financial applications. However, due to the intellectual property of model weights and the copyright of training corpus and benchmarking questions, verifying the legitimacy of GPT's model weights and the credibility of model outputs is a pressing challenge. In this paper, we introduce a novel zkFinGPT scheme that applies zero-knowledge proofs (ZKPs) to high-value financial use cases, enabling verification while protecting data privacy. We describe how zkFinGPT will be applied to three financial use cases. Our experiments on two existing packages reveal that zkFinGPT introduces substantial computational overhead that hinders its real-world adoption. E.g., for LLama3-8B model, it generates a commitment file of $7.97$MB using $531$ seconds, and takes $620$ seconds to prove and $2.36$ seconds to verify.