Exploring Fact Memorization and Style Imitation in LLMs Using QLoRA: An Experimental Study and Quality Assessment Methods

TL;DR

This paper investigates the use of QLoRA, a PEFT method, to adapt large language models for fact memorization and style imitation, evaluating the quality of simulated responses against human interviews.

Contribution

It provides an experimental analysis of QLoRA's effectiveness in style imitation and fact retention in LLMs, along with new assessment methods for response quality.

Findings

QLoRA effectively adapts LLMs for style imitation.

The quality of fact memorization varies with adaptation parameters.

Assessment methods can reliably compare human and model responses.

Abstract

There are various methods for adapting LLMs to different domains. The most common methods are prompting, finetuning, and RAG. In this work, we explore the possibility of adapting a model using one of the PEFT methods - QLoRA. The experiment aims to simulate human responses based on their interviews. The simulation quality is assessed by comparing the quality of the style and the quality of the generated facts.