AI Glossary

LoRA

Also called low-rank adaptation

What does “LoRA” mean?

LoRA: A fine-tuning technique that freezes the original model weights and trains small add-on matrices instead. Only a tiny fraction of the parameters change, so it needs much less memory, and the resulting adapter files are small enough to swap in and out easily.

LoRA visual guide: A fine-tuning technique that freezes the original model weights and trains small add-on matrices instead. Only a tiny fraction of the parameters change, so it needs much less memory, and the resulting adapter files are small enough to swap in and out easily.
Visual guide Open full size (new tab)

What is an example of “LoRA”?

Image-generation communities widely share LoRA adapters that teach a base model a particular art style or character without retraining the whole model.

Keep learning

See also

QLoRATermFine-tuningTermLoRA (image models)TermQuantizationTerm

Same topic

Loss functionTermLearning rateTermOverfittingTermKnowledge distillationTermPre-trainingTerm