1 result for “LLM adaptation”
LoRA is a parameter-efficient fine-tuning technique that adapts large pre-trained models by injecting small trainable low-rank matrices into transformer layers, drastically reducing the number of trainable parameters without sacrificing performance.