ImportError: cannot import name 'AutoModel' from 'transformers'

🐛 Bug

(Not sure that it is a bug, but it is too easy to reproduce I think)

Information

I couldn’t run python -c 'from transformers import AutoModel', instead getting the error on the titile.

To reproduce

Steps to reproduce the behavior:

1. $ sudo docker run -it --rm python:3.6 bash
2. # pip install tensorflow==2.0 transformers==2.8.0
3. # python -c 'from transformers import AutoModel'

Traceback (most recent call last):
  File "<string>", line 1, in <module>
ImportError: cannot import name 'AutoModel'

Initially I got this error with transformers-cli download :

# transformers-cli download t5-large
Traceback (most recent call last):
  File "/usr/local/bin/transformers-cli", line 32, in <module>
    service.run()
  File "/usr/local/lib/python3.6/site-packages/transformers/commands/download.py", line 29, in run
    from transformers import AutoModel, AutoTokenizer
ImportError: cannot import name 'AutoModel'

Expected behavior

no import error.

Environment info

• transformers version: 2.8.0
• Platform: Linux-4.15.0-99-generic-x86_64-with-debian-10.3
• Python version: 3.6.10
• PyTorch version (GPU?): not installed (NA)
• Tensorflow version (GPU?): 2.0.0 (False)
• Using GPU in script?: no
• Using distributed or parallel set-up in script?: no

Also tested on 3.6@Ubuntu 18.04, 3.8@Docker(python:3.8). The Docker backend is Ubuntu 18.04.

Thank you.

[julien-c]

AutoModel is the equivalent of TFAutoModel but for PyTorch model classes. If you don’t have pytorch installed this is expected.

Use TFAutoModel instead =)

[kennylajara]

EDIT: Well… This issue has been solved by updating dependencies but now it produces another issue. I will post the link to the reproducible issue in a minute.