Open BitCalSaul opened 1 month ago
If an user wants to use this method, would he download compressed files by Transformer or the whole Transformer model?
BitCalSaul
Hi. There's no need to transfer the model, since our transformer at the beginning is random initialized.
Hi, thanks for your work, which is interesting. I have a question for the figure below: did you consider the model size when calculating the compression ratio since Transformer model is quite big.