Tag
This paper explores using parameter-efficient fine-tuning (PEFT) as a compact substrate for persistent personal models, studying scaling up, down, and out, and introduces MinT for managing adapters.