LoRA (Low-Rank Adaptation) is a popular technique to finetune LLMs more efficiently. This Studio explains how LoRA works by coding it from scratch, which is an excellent exercise for looking under the hood of an algorithm.