The goal of this work is to enhance implicit fine-grained knowledge transferring, offering the best trade-off between performance and parameter efficiency. If you find this code useful for your ...