How to merge two trained model into one model?

[sigit-purnomo]

Hi all,

I have trained a POS Tagging and Dependency model using the spaCy project template and made the model into a package. I have also trained a model for NER (only the NER component) using a different corpus/dataset and made it into a package.

I have read this thread, but there are many ways. What's the best practice to merge both models into one package?

Thank you

[danieldk]

The standard way we do this is using the spaCy assemble subcommand. assemble allows you to create a spaCy pipeline without training. You can source the components from the pipelines that you already created. You can find an example of sourcing in the coref example project:

https://github.com/explosion/projects/blob/e24a085669b4db6918ffeb2752846089d8dee57a/experimental/coref/configs/coref.cfg#L28

You can then call spacy assemble on the configuration, see the same project for an example.

[sigit-purnomo]

Thank you Daniel