Pipiline for BERT

I`m trying to check how BERT models can improve intent detection for my data and searched all possible info how to use it in Rasa. First of all I found this article How to Use BERT in Rasa NLU and video “Diet Architecture 3: Benchmarking” with examples of config.yml.

The problem is that both examples use deprecated HFTransformersNLP and parameters. Is there are recent, actual examples of what pipelines should be used in Rasa with BERT?

you can replace HFT with Language Model Featurizer as explained in the docs above