Learning to Reduce: Optimal Representations of Structured Data in Prompting Large Language Models
This addresses a bottleneck in using LLMs for tasks involving structured data, offering an incremental improvement over existing methods.
The paper tackles the challenge of integrating structured data into prompts for large language models (LLMs) by proposing a framework that fine-tunes a model to generate reduced versions of input contexts, improving LLM reasoning performance, especially with long contexts, and achieving comparable accuracy in evidence selection across datasets.
Large Language Models (LLMs) have been widely used as general-purpose AI agents showing comparable performance on many downstream tasks. However, existing work shows that it is challenging for LLMs to integrate structured data (e.g. KG, tables, DBs) into their prompts; LLMs need to either understand long text data or select the most relevant evidence prior to inference, and both approaches are not trivial. In this paper, we propose a framework, Learning to Reduce, that fine-tunes a language model to generate a reduced version of an input context, given a task description and context input. The model learns to reduce the input context using On-Policy Reinforcement Learning and aims to improve the reasoning performance of a fixed LLM. Experimental results illustrate that our model not only achieves comparable accuracies in selecting the relevant evidence from an input context, but also shows generalizability on different datasets. We further show that our model helps improve the LLM's performance on downstream tasks especially when the context is long.