Update quickstart notebooks on website

[calpt]

Required updates for quickstart training & inference notebooks linked in the homepage:

• use stable version of adapter-transformers from PyPI
• avoid cloning HF transformers (confusing for users)
• switch from HF transformers GLUE processing (deprecated) to HF datasets

[arueckle]

Any update on this?

[arueckle]

Remove the tipp on

Tip 2️: Depending on your data set size you might also need to train longer than usual. To avoid overfitting you can evaluating the adapters after each epoch on the development set and only save the best model.

We found no evidence for this (see adapterdrop paper)

[calpt]

Remove the tipp on

Tip 2️: Depending on your data set size you might also need to train longer than usual. To avoid overfitting you can evaluating the adapters after each epoch on the development set and only save the best model.

We found no evidence for this (see adapterdrop paper)

Isn't it still true that you would train for more epochs usually? (I think what's meant in this tip is to train for more epochs and not necessarily longer in training time as adapter training is faster).