Ask for help，I'm having a problem.

[sdlmw]

I used the NMT tool to training a Japanese to English engine，Cleaned up unnecessary impurities. It's always a bad result.I want to know if this is a bug that the software learns from Double-byte text，Or I'm using a nonstandard parameter（-layers 3 -rnn_size 500 ）.Has anyone ever encountered the same problem as me?

thanks everybody

[sdlmw]

problem：
1.Not translated
2.Duplicate text
3.Missing words

thanks

[guillaumekln]

How large is the training dataset you are using? How did you prepare/tokenize it?

[sdlmw]

@guillaumekln yeah， about 4 million。I have used the Mecab tool to do the tokenize

[qutie75]

Hello! @sdlmw
I am training a Korean-English model now, and using -layers 8 -rnn_size 1000 options.
My data size is about 3 million and I also used the Mecab for tokenizing.
I am not so sure how you clean up the unnecessary things, but I don't have any problem about Double-byte text until now.