Fine-tuning & LoRA ☆

Fine-tuning continues training on a narrow dataset to specialize a model; LoRA does it by learning tiny low-rank adapter matrices instead of touching all weights, cutting cost by orders of magnitude. It is how enterprises adapt open models, how the Stable Diffusion style ecosystem works, and what products like Tinker sell as a service.