Huggingface warnings cleanup

[radswn]

I've stumbled upon this issue that might get rid of that weird Some weights of the model checkpoint at bert-base-uncased were not used when initializing Bert... warnings that you probably noticed every time you run any notebook:)

The thing is, solution proposed there to simply increase logs verbosity level - transformers.logging.set_verbosity_error() does not necessarily align with the guidlines from our supervisor about the amount of logs. Would be nice to hear your take, whether it's worth pursuing - just one (two, if you count the import) line, but a general limitation to our logs as well:/

[radswn]

An extension of #32