Offsite-Tuning: Transfer Learning without Full Model

Guangxuan Xiao, Ji Lin, Song Han

Transfer learning is important for foundation models to adapt to downstream tasks. However, many foundation models are proprietary, so users must share their data with model owners…