lang.from_bytes does not recreate the pipeline properly

[oliviercwa]

I need to load and save pipeline as bytes. When using the following example, the pipeline that is being loaded does not have its pipeline components set.

What is the right way of saving a custom pipeline to a stream and loading it back?

Thank you,
Olivier

How to reproduce the behaviour

nlp = spacy.load('en_core_web_md')
with open('c\\test', 'wb') as outfile:
  outfile.write(nlp.to_bytes())

# This one will detect entities properly
test = nlp('I am going to Germany')

nlp = spacy.blank('en')
with open('c:\\test', 'rb') as outfile:
  nlp.from_bytes(outfile.read())

# NER has not been loaded.
test = nlp('I am going to Germany')

Your Environment

• Operating System: Windows 10
• Python Version Used: 3.7.4
• spaCy Version Used: 2.1.9
• Environment Information:

[adrianeboyd]

See: https://spacy.io/usage/saving-loading#pipeline