Quantum-PEFT: Ultra parameter-efficient fine-tuning
This addresses the bottleneck of parameter efficiency in fine-tuning large models, offering a novel quantum-inspired approach that could benefit AI practitioners dealing with high-dimensional data.
This paper tackles the problem of parameter-efficient fine-tuning (PEFT) by introducing Quantum-PEFT, which uses quantum unitary parameterization to reduce trainable parameters logarithmically with dimension, achieving vanishingly smaller parameters than LoRA while maintaining competitive performance on language and vision benchmarks.
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 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. Quantum-PEFT achieves 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.