Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time

Wortsman, Mitchell, Ilharco, Gabriel, Gadre, Samir Yitzhak · arXiv · 2022