Columbia University / NYU · Finance
FinGPT
An open-source financial language model framework that democratizes financial AI through data-centric approaches and lightweight fine-tuning.
Overview
FinGPT is an open-source financial LLM framework that takes a data-centric approach to building financial AI. Rather than training a massive model from scratch, FinGPT provides tools for curating financial data from diverse sources and efficiently fine-tuning existing open-source LLMs for financial tasks. The project includes automated financial data pipelines, LoRA-based fine-tuning recipes, and benchmarks, making it a practical toolkit for financial NLP research and applications.
Framework
Data-centric LLM fine-tuning toolkit
Base Models
LLaMA, ChatGLM, Falcon, and others
Fine-Tuning
LoRA / QLoRA
Data Sources
SEC filings, news, social media, market data
License
MIT
Capabilities
Financial sentiment analysis
Stock movement prediction
Financial report analysis
Robo-advisory text generation
Financial data pipeline automation
Use Cases
Building custom financial chatbots with up-to-date market data
Fine-tuning LLMs for firm-specific financial analysis needs
Predicting stock price movements from news and filings
Creating automated financial advisory content
Pros
- +Fully open-source with MIT license for commercial use
- +Data-centric approach enables customization with fresh data
- +Supports multiple base models for flexibility
- +Lightweight fine-tuning accessible to smaller organizations
Cons
- -Requires technical expertise for data pipeline setup
- -Performance depends heavily on the quality of curated data
- -Not a single pre-trained model but a framework requiring assembly
- -Benchmark results vary by base model and fine-tuning quality
Pricing
Free and open-source. Fine-tuning costs vary based on base model size and compute; LoRA fine-tuning can be done on a single consumer GPU.