I'm trying to shift over to Spacy 3.0's training config file framework and am having trouble adjusting the settings to what I'd like to do. Simply put, I would like to use one of the out of the box textcat models (say, bag of words), but pass in the word embeddings produced by a pretrained transformer (e.g., bert base cased), without any fine tuning. So far I've been working off of the texcat config template provided on the Spacy website.
Any help would be much appreciated. I can provide additional details if necessary. Thank you!