how to train an NEL model with transformers?

[moyid]

Hi I trained an NER model with transformers, and now I want to add entity linking to it. The problem is that the NER model I trained doesn't return any vector representations of text, and for NEL I need to create a knowledge base, where you need to add the entity_vector.

when I run this code:

vectors_model = "path/to/transformers/NER"
nlp = spacy.load(Path(vectors_model))
kb = KnowledgeBase(vocab=nlp.vocab, entity_vector_length=300)

for qid, names in name_dict.items():
    desc = ' '.join(names)
    desc_doc = nlp(desc)
    desc_enc = desc_doc.vector
    kb.add_entity(entity=qid, entity_vector=desc_enc, freq=342)

I get the error:
ValueError: [E141] Entity vectors should be of length 300 instead of the provided 0.

[moyid]

following up on this, any suggestions? Thank you!

[svlandeg]

Hi @moyid, a related discussion thread is here: #6511. That thread discusses how to set up user hooks to derive the vector property from the tensors that are being produced by the transformer in your pipeline. Hope that helps!

[nkaramooz]

@moyid were you able to get this working? I'm still struggling despite reviewing the related thread.