lora

LoRA stands for *Low-Rank Adaptation*. It is a machine learning technique used to fine-tune large language models (and other neural networks) more efficiently. Instead of updating all the parameters of a model, LoRA introduces small, trainable rank decomposition matrices into certain layers, significantly reducing the number of trainable parameters while maintaining model performance.