Tag: parameter-efficient fine-tuning

LoRA fine-tuning from the mechanic up: the low-rank update, the zero initialisation, rank and alpha, the honest comparison, and merging versus adapter…