How to make custom trained spaCy pipeline more portable?

[funghetto]

I created a pipeline for my needs following the documentation, however the folder it creates is over 600mb and when I distribute the code to run it it requires the inclusion of spacy-transformers as a runtime dependency.
This is not the case with the official models that are like 20mb and don't require that heavy dependency, what steps do I need to take to make my pipeline as light as the official ones?

[polm]

Some of the pretrained pipelines are small because they don't use transformers. The pretrained transformers pipelines, like en_core_web_trf, are over 400MB in size - transformers are just big. If you train a model without transformers it should be much smaller, though it depends on your use of word vectors, for example.