Inference-Time Intervention in Large Language Models for Reliable Requirement Verification
This work addresses the problem of reliable requirement verification for engineers in systems engineering, offering a dynamic control method that is incremental over existing fine-tuning and prompting approaches.
The paper tackled the challenge of steering Large Language Models (LLMs) for precise control in engineering applications by using inference-time intervention techniques to automate requirement verification in Model-Based Systems Engineering (MBSE). The result was robust and reliable outputs, achieving perfect precision on a holdout test set when combined with self-consistency.
Steering the behavior of Large Language Models (LLMs) remains a challenge, particularly in engineering applications where precision and reliability are critical. While fine-tuning and prompting methods can modify model behavior, they lack the dynamic and exact control necessary for engineering applications. Inference-time intervention techniques provide a promising alternative, allowing targeted adjustments to LLM outputs. In this work, we demonstrate how interventions enable fine-grained control for automating the usually time-intensive requirement verification process in Model-Based Systems Engineering (MBSE). Using two early-stage Capella SysML models of space missions with associated requirements, we apply the intervened LLMs to reason over a graph representation of the model to determine whether a requirement is fulfilled. Our method achieves robust and reliable outputs, significantly improving over both a baseline model and a fine-tuning approach. By identifying and modifying as few as one to three specialised attention heads, we can significantly change the model's behavior. When combined with self-consistency, this allows us to achieve perfect precision on our holdout test set.