cbaziotis / lm-prior-for-nmt Goto Github PK

Hi, Thanks for making this repo public. Could you please add scripts for generating the translations to the repository i.e. the evaluation and inference scripts as it is not present in the current repository?

Hi! I have tried to reproduce your work (en->tr direction transformer baseline for now) using both fairseq and your system. I followed the exact data preprocessing and hyper-parameter settings in your configs. And I only achieved 9.7 sacrebleu score instead of 11.2. I would appreciate it if could you please double-check the code or hyper-parameter settings for that!

Below is the result I got. PS: I checked the sacrebleu version in your requirements.txt as well, it yields the same score.

{
 "name": "BLEU",
 "score": 9.7,
 "signature": "nrefs:1|case:mixed|eff:no|tok:13a|smooth:exp|version:2.3.1",
 "verbose_score": "38.5/13.9/6.0/2.7 (BP = 1.000 ratio = 1.042 hyp_len = 55984 ref_len = 53731)",
 "nrefs": "1",
 "case": "mixed",
 "eff": "no",
 "tok": "13a",
 "smooth": "exp",
 "version": "2.3.1"
}

Thanks for your great work!
I am wondering where is the "mono_sent.py"(the file used for training the language model)?

Looking forward to your reply!

Hi,

I am trying to train a TM using the following command:
python nmt_prior.py --config ../configs/transformer/trans.deen_base.yaml --name final.trans.deen_base
The training was good until the end of the last epoch, an error appeared:

Traceback (most recent call last):
File "/home/ubuntu-user/mc35180/lm-prior-for-nmt/models/nmt_prior.py", line 143, in
eval_best(trained_model)
File "/home/ubuntu-user/mc35180/lm-prior-for-nmt/models/../models/nmt_prior_helpers.py", line 121, in eval_best
eval_nmt_checkpoint(trainer.best_checkpoint,
File "/home/ubuntu-user/mc35180/lm-prior-for-nmt/models/../models/translate.py", line 110, in eval_nmt_checkpoint
_base, _file = os.path.split(checkpoint)
File "/home/ubuntu-user/anaconda3/envs/lmprior/lib/python3.9/posixpath.py", line 103, in split
p = os.fspath(p)
TypeError: expected str, bytes or os.PathLike object, not NoneType

And there is no any checkpoints ever saved in the experiment directory. How can I fix this?

Hi! I'm trying to reproduce your en-tr nmt base model, and I just found it needs 200,000 epochs to train in configs/transformer)/trans.entr_base.yaml. Do you probably have any ideas about how much time will this take (only the nmt base model for en->tr)?

when i run your code,i meet the above error.Can you help me to solve it?Thanks a lot.

Fabulous work and paper！
I'd like to train a EN-DE translation model based on your code. First, I train a DE LM, Then I train a EN-DE TM with LM prior,but I got BLEU 0.0 on val and test, is there something wrong I have made? My environment is same as yours.

Thanks for your patience!!

Great works! Your findings are really appreciable.

When performing the data preprocessing by myself, I found that the "tools" directory is missing in this repository. It will be very helpful if you could upload that too?

Please find the screenshot of preprocessing shell script where you have written code about the same.