Vector-Based Adaptation

A parameter-efficient fine-tuning approach that adapts models using learned vectors instead of full weight matrices, requiring even fewer parameters than LoRA.