LoRA Learns Less and Forgets Less

preview_player
Показать описание
LoRA is a parameter-efficient finetuning method for large language models, but underperforms full finetuning in most cases. It offers strong regularization and diverse generations.

Рекомендации по теме