Ahmedd-Wahdan

why in transformer we compute for all tokens but then use only the last token for prediction?

3

the input is (B,T) to the transformer and the output from the MLP is also (B,T) and we only use the embeddings of the last column to predict the next...