Syrup274

I was using seq_out which does not contain any extra dense layer, but i still got different output with the same input when I load weights from given checkpoint. Is...

I truned off the dropout, set is_training=False and is_finetune=True, but get different results for seq_out. Did you encounter the same problem?