Quantum-PEFT: Ultra parameter-efficient fine-tuning

Toshiaki Koike-Akino, Francesco Tonin, Yongtao Wu, Leyla Naz Candogan, Volkan Cevher

Keywords: parameter-efficient fine-tuning, lora, quantum machine learning, orthogonality constraints

TL;DR: We achieve ultra parameter-efficient fine-tuning by parameterizing the low-rank subspaces via Kronecker products of generalized Pauli rotations

Abstract: This paper introduces Quantum-PEFT that leverages quantum computations for parameter-efficient fine-tuning (PEFT). Unlike other additive PEFT methods, such as low-rank adaptation (LoRA), Quantum-PEFT exploits an underlying full-rank yet surprisingly parameter-efficient quantum unitary parameterization with alternating entanglement. With the use of Pauli parameterization, the number of trainable parameters grows only logarithmically with the ambient dimension, as opposed to linearly as in LoRA-based PEFT methods. Consequently, Quantum-PEFT achieves a vanishingly smaller number of trainable parameters than the lowest-rank LoRA as dimensions grow, enhancing parameter efficiency while maintaining a competitive performance. We apply Quantum-PEFT to several transfer learning benchmarks in language and vision, demonstrating significant advantages in parameter efficiency.

Submission Number: 8