In this PR I push a little fix to disable the validation loop. As we discussed in the last meeting, the parser is not able to differentiate between NanosetDatasetsArgs & MultilingualNanosetDatasetsArgs properly.

You can deactivate the validation loop by NOT setting validation_folder or by manually setting val_check_interval = -1.

The parser is a bit stupid and won't be able to parse properly as MultilingualNanosetDatasetsArgs when specifying training_folder & validation_folder, so we will never raise this error. Instead, it will raise the typical parser error which looks like this:

dacite.exceptions.UnionMatchError: can not match type "dict" to any type of "data_stages.data.dataset" union: typing.Union[nanotron.config.config.PretrainDatasetsArgs, nanotron.config.config.MultilingualNanosetDatasetsArgs]

Nevertheless, despite raising the parser error instead of the other one complaining about the missing languages the functionality is the desired one.

PD: Why I'm pushing so many commits? Aren't merged yet? Do they come from the future?