Fine-tuning Smaller Language Models for Question Answering over Financial Documents

TL;DR

This paper demonstrates that smaller language models, when fine-tuned with reasoning exemplars, can effectively perform multi-hop financial reasoning tasks, approaching the performance of larger teacher models.

Contribution

It introduces a fine-tuning approach for smaller models in financial question answering, emphasizing multi-hop reasoning and demonstrating comparable performance with less data.

Findings

Fine-tuned smaller models approach teacher model performance.

Fine-tuning improves financial concept expression and entity extraction.

Comparable reasoning achieved with smaller datasets.

Abstract

Recent research has shown that smaller language models can acquire substantial reasoning abilities when fine-tuned with reasoning exemplars crafted by a significantly larger teacher model. We explore this paradigm for the financial domain, focusing on the challenge of answering questions that require multi-hop numerical reasoning over financial texts. We assess the performance of several smaller models that have been fine-tuned to generate programs that encode the required financial reasoning and calculations. Our findings demonstrate that these fine-tuned smaller models approach the performance of the teacher model. To provide a granular analysis of model performance, we propose an approach to investigate the specific student model capabilities that are enhanced by fine-tuning. Our empirical analysis indicates that fine-tuning refines the student models ability to express and apply the required financial concepts along with adapting the entity extraction for the specific data format. In addition, we hypothesize and demonstrate that comparable financial reasoning capability can be induced using relatively smaller datasets.