InvalidArgumentError: logits and labels must have the same first dimension, got logits shape [32,32] and labels shape [64]

Question

I am trying to convert the single-word answers question-answer model to multiple-word answers. I am using the Keras implementation of dynamic memory nets. The code I updated