Inputs to lstm layer:

in the case of Neural Machine Translation (NMT) model to translate human-readable dates (β€œ25th of June, 2009”) into machine-readable dates (β€œ2009-06-25”).the post-attention LSTM at time 𝑑 does not take the previous time step’s prediction π‘¦βŸ¨π‘‘βˆ’1⟩ as input
if we require an attention model for which the post lstm layer should have inputs both y(t-1) (the previous output) and the context vector at that time step(for example in case of machine translation)
can we just concatenate both y(t-1) and context vectors or can we give two inputs to the lstm layer by using the call arguments β€œinputs”

Were you able to find an answer for your question?