search

facebookresearch / mlqe

We release a dataset based on Wikipedia sentences and the corresponding translations in 6 different languages along with the scores (scale 1 to 100) generated though human evaluations that represent the quality of the translations.Paper Title Unsupervised Quality Estimation for Neural Machine Translation
Creative Commons Attribution Share Alike 4.0 International
80 stars 14 forks source link

More details on models? #4

Closed shuoyangd closed 3 years ago

shuoyangd commented 3 years ago

Hi,

I'm trying to reproduce the models that are provided as part of the shared task. However, I found the hyper-parameter settings for these models were only vaguely described in the paper (was looking at this one).

Is it possible to provide more details on that? Most desirably, the exact fairseq command used to train these models will be very helpful.

Thanks!

Shuoyang

shuoyangd commented 3 years ago

Ah, just realized that I can recover these arguments from the model dump. The model dump is a dict with an entry args.