Fine-Tuning Large Language Models for Quantum Reasoning

2026-06-20

AI Summary

This work investigates how to endow large language models with genuine quantum reasoning capabilities—beyond mere task-specific pattern matching—by fine-tuning them to simulate quantum circuits. Specifically, the models are trained to predict the measurement probability distributions generated by sequences of quantum gates. Two fine-tuning strategies are proposed: supervised fine-tuning (SFT) using explicit state-vector trajectories, and a two-stage approach that augments SFT with Group Relative Policy Optimization (GRPO). Experimental results demonstrate that SFT achieves near-perfect accuracy on both in-distribution tasks and gate-count extrapolation, substantially outperforming baseline methods. Although the SFT+GRPO variant incurs a slight reduction in accuracy, it significantly enhances generalization to larger-scale qubit systems.

Abstract

Large language models (LLMs) exhibit abilities beyond natural language modelling and text generation. Recent advances in their reasoning capabilities have spurred interest in applying LLMs to complex scientific tasks requiring deep domain expertise and sophisticated reasoning. Quantum computing, as a highly specialised field with significant knowledge barriers and hardware constraints, could greatly benefit from such advancements. However, a key open question that first must be answered is: How can we develop fine-tuning pipelines that instil genuine quantum reasoning in LLMs, rather than task-specific pattern matching? We study this question through quantum circuit simulation as a training objective, where the model must predict the measurement probability distribution resulting from a sequence of quantum gate operations. We propose and compare two fine-tuning pipelines: (1) Supervised Fine-Tuning (SFT) on explicit gate-by-gate state-vector simulation traces, and (2) a two-stage SFT+Group Relative Policy Optimisation (GRPO) approach that sequentially applies SFT followed by GRPO with verifiable rewards. Our findings show that SFT achieves near-perfect in-distribution and gate-count extrapolation accuracy, significantly outperforming both the base model and the GPT-OSS-120B baseline. SFT+GRPO trades some in-distribution precision for better generalisation to larger qubit systems that SFT alone cannot handle. Both pipelines significantly outperform the baselines, demonstrating that targeted fine-tuning on explicit reasoning traces is an effective strategy for advancing quantum reasoning in LLMs.

Problem

• quantum reasoning
• large language models
• fine-tuning
• quantum computing
• reasoning capability

Innovation

• quantum reasoning
• supervised fine-tuning
• group relative policy optimisation
• quantum circuit simulation
• large language models