Masking input timesteps for LSTMBuilder.add_inputs?

Hi all,

I am currently manually minibatching my code for a seq2seq model. I am using masking to batch together my examples. I understand that for the output, I can multiply each output token's loss by the mask value, and have a masked output.

How do I mask the input? I think that set_dropout_masks seems to be the solution to my problem, but I'm not sure how to use it. Any help would be greatly appreciated! Thanks!