THaLLE: Text Hyperlocally Augmented Large Language Extension -- Technical Report

TL;DR

This paper introduces THaLLE, an 8-billion-parameter LLM fine-tuned for financial analysis, achieving top performance on CFA mock exams and providing a dataset for evaluating financial advising capabilities.

Contribution

The work presents a specialized financial extension of a hyperlocally augmented LLM, with detailed fine-tuning methods and a new dataset for financial evaluation.

Findings

THaLLE outperforms comparable models on CFA mock exams

Fine-tuning techniques enhance financial reasoning in LLMs

Introduces Flare CFA dataset for financial model evaluation

Abstract

Recent advancements in Large Language Models (LLMs) have revealed new capabilities and opportunities across the technological landscape. However, the practicality of very large LLMs is challenged by their high compute cost, which does not justify the benefits given their limited capability compared to humans. While smaller, more practical LLMs have shown potential in financial analysis, though they are not yet fully proficient, as evidenced by their near-passing performance on the Chartered Financial Analyst (CFA) exam. In this work, we present Financial Analyst Extension to our Text Hyperlocally Augmented Large Language Extension (THaLLE), a series of 8B LLMs consistently achieving highest performance on mock CFA exams against models of comparable size. We thoroughly document the fine-tuning techniques used to facilitate future research. Additionally, we introduce the use of Flare CFA, a publicly available dataset for evaluating LLMs as a financial advisor.