ModuleNotFoundError - "model" directory is missed in setup.py

Bug description “model” directory is missed in “packages” argument in setup.py script. That issue raised ModuleNotFoundError: No module named 'esm.model' exception.

Reproduction steps Try to install esm package from PyPI or from the GitHub repository.

Expected behavior I was expected successful installation.

Logs

---------------------------------------------------------------------------
ModuleNotFoundError                       Traceback (most recent call last)
Input In [1], in <cell line: 2>()
      1 import torch
----> 2 import esm

File ~/protein_embeddings/env/lib/python3.8/site-packages/esm/__init__.py:9, in <module>
      6 from .version import version as __version__  # noqa
      8 from .data import Alphabet, BatchConverter, FastaBatchedDataset  # noqa
----> 9 from .model.esm1 import ProteinBertModel  # noqa
     10 from .model.esm2 import ESM2  # noqa
     11 from .model.msa_transformer import MSATransformer  #noqa

ModuleNotFoundError: No module named 'esm.model'

[ptynecki]

@tomsercu PR #254 is resolving the issue problem but there is another during the installation.

The fairscale package is required (not added to setup.py as well).

[matteoferla]

I might be wrong, but the setup.py is rather curiously written, I don’t know your guys setup and I have not looked at the history, but it very much looks like be a case of files being erroneously switched for a test or similar. Say

• the setuptools.setup function lacks a requirement argument (i.e. no fairscale module))
• setuptools.find_packages fills subpackages without issues (okay, PyCharm does it but is bad)
• reading files in the setup.py file is bad and may raise a FileNotFoundError if packaged as a wheel without the Manifest.in file
• __version__ is read very oddly
• etc.

So might be worth checking there’s no weird accident in the version!