Abstract

This research introduces a hybrid classical-quantum framework for textclassification, integrating GPT-Neo 125M with Low-Rank Adaptation (LoRA) andSynthetic Minority Over-sampling Technique (SMOTE) using quantum computingbackends. While the GPT-Neo 125M baseline remains the best-performing model,the implementation of LoRA and SMOTE enhances the hybrid model, resulting inimproved accuracy, faster convergence, and better generalization. Experimentson IBM's 127-qubit quantum backend and Pennylane's 32-qubit simulationdemonstrate the viability of combining classical neural networks with quantumcircuits. This framework underscores the potential of hybrid architectures foradvancing natural language processing applications.

 

Quick Read (beta)