[SOLVED] ValueError: [E002] Can't find factory for 'transformer' for language Italian (it)

[Panzerotto93]

How to reproduce the behaviour

Create docker image with:

• spacy==3.2.4
• spacy[transformers]==3.2.4
• transformers==4.17.0
• tokenizers==0.12.1
  In requirements. I downloaded the model files and put them in a folder inside the docker image.
  Calling spacy.load(path/to/model/it_nerIta_trf), will raise the error in the title.

Your Environment

• spaCy version: 3.2.4
• Platform: Windows-10-10.0.19044-SP0
• Python version: 3.9.7
• Pipelines: it_nerIta_trf (0.0.0)

How do I solve this? I have no problems running the code locally on Windows, but the error is raised in a Linux Docker container. Thanks!

[polm]

Have you installed spacy-transformers? Not finding the transformer factory doesn't mean the model files can't be found, it means spaCy doesn't have any knowledge of Transformer functionality. If you install a model through pip then dependencies like spacy-transformers will be automatically installed, but if you call spacy.load that won't happen, so you need to insure you already have dependencies.

[Panzerotto93]

Guess the problem is with importlib.metadata. Running

import importlib.metadata as importlib_metadata
importlib_metadata.version('tqdm')

raises ModuleNotFoundError
tried with:

import importlib
importlib.sys.path.append('/usr/local/lib/python3.8/site-packages/')
import importlib.metadata as importlib_metadata
importlib_metadata.version('tqdm')

Same thing happens

[polm]

We have had weird problems with importlib before, but nothing like that, and I'm honestly not sure what could cause it.

Kind of a long shot but maybe you could try changing your Python version to 3.9? Since importlib is a core library that might change something subtle in it.

Besides that... I'm not sure I can offer much more help unless you can make an example we can reproduce somehow to investigate. It sounds like somthing is wrong with your virtualenv but hard to say what.

[Panzerotto93]

Problem was with the default configuration of docker image made by IT, in which all the dist-info folders were deleted. Thus, every time importlib searched for tqdm dist-info folder, it couldn't find it and raised the error. Just had to comment that line in the dockerfile to solve the problem.
Sorry for bothering you and thanks a lor for the help!

[polm]

Glad you figured it out, and thanks for letting us know what caused it!

[polm]

Also as a note, in this forum we don't edit the titles of threads to add [SOLVED] or similar, we just use the "mark answer" feature to mark something as "done". Sometimes it's a little weird (like when the first post isn't a solution), but that's OK, as is not marking the thread solved in any way - we also have an internal tracker.