The results of the translate function.

[zshyang]

I have trained the model for WMT'16. And the accuracy and loss are similar to what they provided.

But when I translate,

python translate.py -data_pkl m30k_deen_shr.pkl -model output/lr_mul_0.5-scale_emb/model.chkpt -output prediction.txt

what I got is the following:

<unk> <unk> their around as as <unk> to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> <unk> all around as as to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> their but as as as to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> their all around as as as to by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> <unk> around as to . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> <unk> as around around as as as as <unk> <unk> by by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> <unk> around as <unk> to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> <unk> quickly around as as as as as <unk> by by by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> <unk> around as to . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> around but as around as to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> their but as as as to by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> <unk> as as around as as as as <unk> by by by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> <unk> around as as to to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> <unk> around around as as as as by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> their around as to . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> <unk> around as as <unk> to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> their around as around to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
<unk> <unk> <unk> but around as as as as <unk> by by . . 

I have no clue what is going on. Could you help me with this please? Thank you.

[nuges01]

Watching. I'm having a similar issue. I don't know if anyone monitors these issues anymore though. Were you able to solve it?

If anyone has any insights on how to get translate working, please let us know.

[eyuansu62]

you can set scale_emb_or_prj=emb, and the result would be better

[1338199]

I met the similar issue, have you solved it?

[HUIYINXUE]

I met the similar issue while trying to reproduce the result on WMT14. Has anyone solved that?