Finance-LLM is an instruction fine-tuned language model built on Llama 2 13B, developed by researchers Zhao and colleagues to advance financial AI reasoning capabilities. It represents one of the early efforts to adapt large language models specifically for financial question answering and reasoning tasks.
The model was trained on a comprehensive financial instruction dataset comprising 150,000 financial Q&A pairs covering corporate finance, investment analysis, financial accounting, and market microstructure. This dataset was carefully curated from financial textbooks, analyst reports, and financial exam preparation materials, providing broad coverage of financial concepts and reasoning patterns.
While Finance-LLM is based on the Llama 2 architecture (which has since been superseded by Llama 3 and later models), it remains a valuable reference for the financial AI research community. Its 150K instruction pairs provide a strong foundation for transfer learning, and the model's performance on financial reasoning benchmarks demonstrates the effectiveness of domain-specific instruction tuning. For researchers and practitioners prototyping financial AI applications, it offers a well-documented starting point for understanding how instruction tuning shapes financial domain capabilities.