Possible issue with validation set creation?

I was giving a good read to https://github.com/argosopentech/argos-train/blob/master/argostrain/data.py#L21 and noticed that the input is not shuffled.

Shouldn't the the input should be shuffled to have a better distribution? For example, if your first source is a wiktionary dataset, the program would be validating the dataset only against single word translations.

Read argostrain/data.py around line 21 and trace how the validation set is created from the input data. Run the repository's existing tests or a small reproduction with an ordered dataset; done means validation data is distributed rather than remaining grouped by the input's original ordering.

Written by the indexing model from the issue text.