PyTorch implementation of low-rank adaptation (LoRA), a parameter-efficient approach to adapt a large pre-trained deep learning model which obtains performance on-par with full model fine-tuning.