Open Source Financial LLMs
Open FinLLMs
Open FinLLMs is a comprehensive initiative developed collaboratively at The Fin AI to advance large language models in financial applications. It focuses on creating open-source financial LLMs, such as FinLLaMA, FinLLaMA-instruct, and FinLLaVA, to enhance accessibility and innovation in financial AI.
Open-FinLLMs is a groundbreaking suite of open-source financial large language models designed to bring institutional-grade analysis to startups, researchers, and emerging markets. By offering cutting-edge financial AI capabilities in an open format, Open-FinLLMs aims to challenge the status quo and democratize access to high-quality financial insights.
Developed through an academic collaboration within The Fin AI community, Open-FinLLMs addresses key gaps in financial AI that proprietary models fail to fill. Unlike closed-source models with limited datasets and no multimodal support, Open-FinLLMs provides an open, extensible solution that significantly advances financial AI. By setting new benchmarks in financial AI performance, Open-FinLLMs empowers researchers and businesses to build more transparent, efficient, and innovative financial tools.
What Open FinLLMs Include?
Built on Meta’s Llama 3 architecture and trained on a groundbreaking 52-billion-token financial corpus, Open-FinLLMs introduces three specialized models:
FinLLaMA: A foundational model pretrained on diverse financial data, including SEC filings, earnings calls, and market indicators.
FinLLaMA-Instruct: Instruction-tuned with 573,000 financial task examples for precise sentiment analysis, risk assessment, and numerical reasoning.
FinLLaVA: The first open-source multimodal financial LLM, capable of interpreting charts, tables, and text simultaneously.
FOUNDATION
FinLLaMA
FinLLaMA is the foundational model of Open-FinLLMs, trained on 52B financial tokens from diverse sources such as SEC filings, earnings calls, and market indicators.
Training Details
Architecture: LLaMA 3
Training Data: 52B financial tokens + 18B general domain tokens
Pre-training Compute: HiPerGator (64 A100 GPUs, 250h)
Capabilities: Market trend analysis, regulatory filings comprehension, and financial text processing
Best in Zero-Shot
FinLLaMA’s strong zero-shot performance enables its deployment in real-world financial scenarios where labeled data is scarce, such as rapid market analysis, fraud detection, regulatory compliance monitoring, and automated financial reporting.
Dominating Few-Shots
FinLLaMA excels in sentiment analysis, financial text classification, and entity recognition with minimal data, enabling real-time insights where labeled data is scarce.
Mastering Trading
FinLLaMA outperforms baselines in trading, achieving higher returns and lower risk across stocks like TSLA and COIN, making it ideal for quantitative trading and portfolio optimization.
CHAT
FinLLaMA-Instruct
FinLLaMA-Instruct is a fine-tuned version of FinLLaMA, trained on 573K financial instruction examples to enhance its reasoning capabilities. It is optimized for sentiment analysis, risk assessment, and numerical reasoning.
Training Details
Architecture: FinLLaMA + Instruction Fine-Tuning
Training Data: 573K financial instruction examples
Capabilities: Sentiment analysis, structured financial reasoning, enhanced numerical inference
Best Performance
FinLLaMA-Instruct outperforms leading financial and general-purpose LLMs, excelling in sentiment analysis, numerical understanding, and structured market prediction. Designed for precision and reliability, it sets a new standard for financial AI applications.
FinLLaVA
FinLLaVA is the first open-source multimodal financial LLM, extending FinLLaMA with 1.43M multimodal instructions. It can interpret charts, tables, and text simultaneously, making it highly effective for financial decision-making and quantitative analysis.
Training Details
Architecture: FinLLaMA-Instruct + Vision Encoder (CLIP)
Training Data: 1.43M multimodal instructions (financial data, charts, tables)
Capabilities: Chart interpretation, financial reporting analysis, multimodal reasoning
Try It Now!
Experience its capabilities firsthand and see how it can enhance financial decision-making. As the first open-source multimodal financial LLM, it enables researchers, startups, and institutions to analyze complex financial data with greater accuracy and efficiency.
Partnerships
We would like to thank our partners NVAITC, for their generous support in making the Open FinLLMs possible.
Open FinLLMs Contributors
The University of Manchester
University of Florida
Columbia University
Yale Univeristy
University of Montreal
Stevens Institute of Technology
Harvard University
Rensselaer Polytechnic Institute
Gustavus Adolphus College
The National Center of Text Mining, UK
Archimedes RC, GR
Join Us
The Fin AI Model Working Group unites researchers, engineers, and financial experts to develop cutting-edge, open financial AI models. Through collaboration, we ensure innovation, reliability, and accessibility in financial AI. We welcome others to join us in shaping the future of financial AI models.