Query

[malikfahadsarwar]

Thanks for sharing the code
I have a question if you can anwer in the training we donot feed the end token embedding as input as end token have to be predicted by model in the output
in the validation since the end token might be produce anywhere[might be different for each seq with in the batch] so let say if the seq produced by the LSTM is 5 tokens means 5 probs where last prob is of end token now in ground truth let say we have 9 tokens then how we supposed to compute the loss pz if you can anser my query