How train PBSMT + NMT

[JxuHenry]

I trained PBSMT and NMT， but I don't no how to train by PBSMT + NMT

[HAOHAOXUEXI5776]

Hello JxuHenry again! Recently I encounter with a problem when training PBSMT. In the file UnsupervisedMT/PBSMT/run.sh, file named $MOSES_PATH/bin/lmplz is used to train a language model both for the SRC and TGT language. However, a error of "Cannot allocate memory for 88976170976 bytes in malloc" for learning English language model occured. The English monolingual corpus contains 10 million sentences. Because the number "10 million" is a default valueI in the original run.sh, I wonder whether you have decrease this number, or 10 million is just ok and won't need that large memory.

[HAOHAOXUEXI5776]

I used 0.1 million sentence in all.en.true to train a language model just now, but the above problem occured again. I guess the problem may lie in my machine.

[HAOHAOXUEXI5776]

ooo, I solved it by adding an argument -S to lmplz. It looks like this:
$TRAIN_LM -o 4 -S 40% < $TGT_TRUE > $TGT_LM_ARPA

[JxuHenry]

ooo, I solved it by adding an argument -S to lmplz. It looks like this:
$TRAIN_LM -o 4 -S 40% < $TGT_TRUE > $TGT_LM_ARPA

Hi, do you know how to slove my problem?

[HAOHAOXUEXI5776]

ooo, I solved it by adding an argument -S to lmplz. It looks like this:
$TRAIN_LM -o 4 -S 40% < $TGT_TRUE > $TGT_LM_ARPA

Hi, do you know how to slove my problem?

Sorry... but I'll share my solution once I get it.

[electleaf]

Were you able to train Nmt+pbsmt ?