Load Custom Local Transform Model in config.cfg

[leoitcode]

Hi people! I'm trying to fine-tune with a portuguese BERT trained in legal documents, but this BERT isn't in huggingface hub. At my config.cfg in portuguese setting generated in spaCy web, I got:

[components.transformer.model] @architectures = "spacy-transformers.TransformerModel.v1" name = "neuralmind/bert-base-portuguese-cased"

This bert-base-portuguese-cased is the default portuguese model on spaCy setting, instead of this I want to load other portuguese custom BERT from local path, and this model isn't in huggingface, but in local with structure:

pytorch_model, vocab.txt, config.json

How could I load this model from local path in config settings?

[adrianeboyd]

Hi, the name can be anything that can be loaded by AutoModel.from_pretrained, including a path to a local directory rather than a model from the huggingface hub.

[leoitcode]

Thanks @adrianeboyd !!!

[leoitcode]

I'm not able to find this information in documentation, but thank you.

[adrianeboyd]

https://spacy.io/api/architectures#TransformerModel

If you have an architecture name like spacy-transformers.TransformerModel.v1 you can usually find the right section in the docs through the search at the top right, where all the config parameters are documented.