What are the science foundation of Spacy NER when I training from scratch?

I am using latest version of spaCy and Prodigy. I annotated Vietnamese for accounting domain from scratch. What are the science foundation of Spacy NER when I training from scratch?

Hi @donhuvy ,

You can train an NER model from scratch by using the train recipe. Under the hood, this calls the spaCy API, so you're virtually training a spaCy NER model. If you're interested in how this process works, you can check out the pipeline design in spaCy.

In addition, once you train a model, a config is generated. You can then check which model is being "accessed" by the config and compare it with the architectures here.