Methods such as adapters, prompt tuning, and low-rank updates keep most pretrained weights fixed. This lowers storage and training requirements when many specialized variants share the same base model.
Parameter-efficient fine-tuning adapts a model by training only a small portion of its parameters or added components.
Methods such as adapters, prompt tuning, and low-rank updates keep most pretrained weights fixed. This lowers storage and training requirements when many specialized variants share the same base model.