# FinGPT: Democratizing Internet-scale Data for Financial Large Language Models

**Authors:** Xiao-Yang Liu, Guoxuan Wang, Hongyang Yang, Daochen Zha  
**arXiv:** [2307.10485](https://arxiv.org/abs/2307.10485)  
**Date:** 2023-07-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2307.10485v2/

## Key points

- Authors propose FinGPT, a data-centric framework for financial large language models (LLMs)
- FinGPT automates collection and curation of real-time financial data from 34 diverse sources
- Introduction of Reinforcement Learning with Stock Prices (RLSP) for fine-tuning FinLLM using market feedback
- Low-rank Adaptation (LoRA, QLoRA) method enables customization of FinLLMs at low cost
- Various applications showcased including robo-advisors, sentiment analysis for algorithmic trading, and low-code development
- Goal of FinGPT is to democratize FinLLMs, foster innovation in open finance, and unlock new opportunities in the field

## Summary

, , , , 

In their paper titled "FinGPT: Democratizing Internet-scale Data for Financial Large Language Models," authors Xiao-Yang Liu, Guoxuan Wang, Hongyang Yang, and Daochen Zha delve into the realm of large language models (LLMs) and their potential impact on the finance industry. The authors discuss the challenges faced by LLMs in understanding and generating financial text due to its unique nature and limited availability of datasets. To address these issues, they propose FinGPT - a data-centric framework that automates the collection and curation of real-time financial data from 34 diverse sources on the Internet. This allows researchers and practitioners to develop their own financial LLMs with accessible resources. The authors also introduce a strategy called Reinforcement Learning with Stock Prices (RLSP) for fine-tuning FinLLM using market feedback. Additionally, they discuss the Low-rank Adaptation (LoRA, QLoRA) method that enables users to customize their own FinLLMs at a low cost from general-purpose LLMs. Various applications of FinGPT are showcased including robo-advisors, sentiment analysis for algorithmic trading, and low-code development. Ultimately, FinGPT aims to democratize FinLLMs, foster innovation in open finance, and unlock new opportunities in the field. The codes for this framework have been made open-source to encourage collaboration and advancement in this domain.