layer sizes to compare iTransformer vs Vanilla Transformers

[akkasayaz]

Hi,

I want to compare iTransformer's encoder only approach to Vanilla Transformer's encoder-decoder type. I used 2 encoder layer for iTransformer and 1 encoder, 1 decoder layer for Transformer with the same d_model and d_ff. However I feel like this is not a fair competition.

AFAIK you published the iTransformer paper so I wanted to ask you.

Thanks for your answer.