LoRA is a parameter-efficient finetuning technique, yes. I recently compared to Llama-Adapter and Llama-Adapter v2:

→ View original post on X — @rasbt