LoRA is an economical way to adapt an AI model: instead of changing all weights, you train only small add-on components.
Example
With a LoRA, an image model learns the style of your company illustrations without the whole model having to be retrained.
For pros
LoRA adds small low-rank matrices to selected weight matrices; only these are trained while the base model stays frozen. That saves memory and compute, and an adapter is only a small fraction of the model's size, often a few to a few hundred megabytes. Variants such as QLoRA combine it with quantization.
