Fine-Tuning Language Models Using Formal Methods Feedback

TL;DR

This paper introduces an automated fine-tuning method for pre-trained language models in autonomous systems, using formal methods feedback to improve domain-specific control policies without human input.

Contribution

It presents a novel automated approach that synthesizes and verifies controllers from language models guided by formal specifications, reducing reliance on human feedback.

Findings

Controller compliance with specifications increased from 60% to 90%.

Method demonstrated effectiveness in autonomous driving tasks.

Automated fine-tuning reduces costs compared to human feedback methods.

Abstract

Although pre-trained language models encode generic knowledge beneficial for planning and control, they may fail to generate appropriate control policies for domain-specific tasks. Existing fine-tuning methods use human feedback to address this limitation, however, sourcing human feedback is labor intensive and costly. We present a fully automated approach to fine-tune pre-trained language models for applications in autonomous systems, bridging the gap between generic knowledge and domain-specific requirements while reducing cost. The method synthesizes automaton-based controllers from pre-trained models guided by natural language task descriptions. These controllers are verifiable against independently provided specifications within a world model, which can be abstract or obtained from a high-fidelity simulator. Controllers with high compliance with the desired specifications receive higher ranks, guiding the iterative fine-tuning process. We provide quantitative evidences, primarily in autonomous driving, to demonstrate the method's effectiveness across multiple tasks. The results indicate an improvement in percentage of specifications satisfied by the controller from 60% to 90%.