SpaCy or BERT for Named Entity Recognition for Metadata extraction for legal documents in the form of text when Training data is just 40 documents

Question

I am working on metadata extraction from documents where I have a training data of just 40 documents and I have to extract information like Aggrement date,