I think my model would benefit from the inclusion of character encoding similar to ELMO, i.e. as discussed in CharacterBERT (https://www.aclweb.org/anthology/2020.coling-main.609.pdf ) since I have loads of acronyms so I think added characters will help.
Is there a layer in the huggingface library I could use to replicate this configuration? I don’t see one, nor a hugginface ELMO but I could well have missed it. Otherwise I’ll try and use AllenNLP.