Translated strings still contain BPE delimiter and EOS token.

[voidmagic]

I am using python 3.6.1 with tensorflow version of 1.2.1.
It seems like the function get_translation didn't work well.
The problem may caused by the reverse_target_vocab_table which convert target sample id into target words.
The result of look up operation on reverse_target_vocab_table is not pure str type. Instead it is something like byte string, which is

[b'<s>', b'</s>']

And this causes mismatch between '</s>' and b'</s>'.