BandSplitRNN-PyTorch
BandSplitRNN-PyTorch copied to clipboard
Does using transformers instead of LSTMs lead to performance degradation?
I noticed that there is a transformer model in the code, it seems that you have considered using it instead of LSTM, I switched it and found that the performance has decreased, is this also the case on your side, what is the reason? Is there a workaround?