Open jindrahelcl opened 6 years ago
In unsupervised scenarios, one usually does not have the validation data at disposal (e.g. in the Unsupervised NMT paper (Artexe et al., 2017))
Enable training for a pre-defined number of steps without validating the model.
For now, you can just set the validation period to a very large number.
... and do not save any trained model.
In unsupervised scenarios, one usually does not have the validation data at disposal (e.g. in the Unsupervised NMT paper (Artexe et al., 2017))
Enable training for a pre-defined number of steps without validating the model.