FinGPT: Open-Source Financial Large Language Models

Hongyang Yang, Xiao-Yang Liu, Christina Dan Wang
FinLLM Workshop @ IJCAI 2023
arXiv: 2306.06031

First official FinGPT paper, accepted by FinLLM Workshop @ IJCAI 2023 and presented in August 2023 (Macao).

Abstract

This paper presents the first official and canonical academic reference for FinGPT, an open-source large language model (LLM) framework specifically designed for the financial domain.

Disambiguation note: This paper is an independent and distinct work and is not a version, extension, or revision of "FinGPT: Democratizing Internet-scale Data for Financial Large Language Models" (arXiv:2307.10485). The latter focuses on instruction data construction (previously referred to as FinNLP) and was accepted later at the Instruction Workshop @ NeurIPS 2023 (October 2023).

Official Academic Statement of FinGPT

FinGPT is an open-source large language model framework for the financial domain, originally proposed and formally defined by the AI4Finance Foundation (first released in June, 2023).

This publication, accepted by the FinLLM Workshop at IJCAI 2023, serves as the first official and canonical academic reference for the FinGPT framework. It introduces an engineering-oriented methodology for building financial LLMs, covering data preparation, model adaptation, and deployment in real-world financial tasks. Since its initial introduction, FinGPT has inspired subsequent research exploring extensions and applications built upon the original framework. Such works should be considered follow-up studies, while this paper remains the original reference for the definition of FinGPT as a framework.

Citation guidance: When referring to FinGPT as a framework or research line, this paper should be cited as the original and canonical reference.

This paper, together with the AI4Finance open-source repositories, constitutes the canonical academic record of the FinGPT project.

Project Stewardship

FinGPT is maintained as an open-source initiative under the AI4Finance Foundation. The official codebase, documentation, and research artifacts are curated by the AI4Finance open-source community.

Open-Source Ecosystem and Research Artifacts

The FinGPT framework is supported by a publicly accessible open-source ecosystem maintained under the AI4Finance Foundation, providing transparent implementations, models, and datasets for research and reproducibility.

• Source Code (GitHub): https://github.com/AI4Finance-Foundation/FinGPT
  Official implementation of the FinGPT framework, including training pipelines, evaluation scripts, and example financial applications.
• Hugging Face Organization: https://huggingface.co/FinGPT
  Official FinGPT organization hosting released models and datasets.
• Pre-trained Models: https://huggingface.co/FinGPT/models
  Financial-domain language models and fine-tuned checkpoints released as part of the FinGPT research project.
• Datasets: https://huggingface.co/FinGPT/datasets
  Curated financial datasets used for model training, instruction construction, and empirical evaluation.

These resources establish a concrete linkage between the original academic publication and its practical open-source instantiations, ensuring transparency and reproducibility of the FinGPT framework.

Full Text

Download PDF

BibTeX

@article{yang2023fingpt_open,
  title   = {FinGPT: Open-Source Financial Large Language Models},
  author  = {Yang, Hongyang and Liu, Xiao-Yang and Wang, Christina Dan},
  journal = {arXiv preprint arXiv:2306.06031},
  year    = {2023},
  note    = {First official FinGPT paper; FinLLM Workshop at IJCAI 2023},
  url     = {https://arxiv.org/abs/2306.06031}
}