name 'PreTrainedTokenizer' is not defined in diffusers 0.22.1

[WiNE-iNEFF]

Describe the bug

when i try to loading pretrained model using DDPMPipeline or DiffusionPipeline, i take error Failed to import diffusers.models.unet_2d because of the following error (look up to see its traceback): name 'PreTrainedTokenizer' is not defined

Reproduction

from diffusers import DiffusionPipeline image_pipe = DiffusionPipeline.from_pretrained(huggan/fastgan-few-shot-painting).to(device)

Logs

NameError                                 Traceback (most recent call last)
/usr/local/lib/python3.10/dist-packages/diffusers/utils/import_utils.py in _get_module(self, module_name)
    683         try:
--> 684             return importlib.import_module("." + module_name, self.__name__)
    685         except Exception as e:

20 frames
NameError: name 'PreTrainedTokenizer' is not defined

The above exception was the direct cause of the following exception:

RuntimeError                              Traceback (most recent call last)
/usr/local/lib/python3.10/dist-packages/diffusers/utils/import_utils.py in _get_module(self, module_name)
    684             return importlib.import_module("." + module_name, self.__name__)
    685         except Exception as e:
--> 686             raise RuntimeError(
    687                 f"Failed to import {self.__name__}.{module_name} because of the following error (look up to see its"
    688                 f" traceback):\n{e}"

RuntimeError: Failed to import diffusers.models.unet_2d because of the following error (look up to see its traceback):
name 'PreTrainedTokenizer' is not defined

System Info

- diffusers` version: 0.22.1

• Platform: Linux-5.15.120+-x86_64-with-glibc2.35
• Python version: 3.10.12
• PyTorch version (GPU?): 2.1.0+cu118 (True)
• Huggingface_hub version: 0.18.0
• Transformers version: not installed
• Accelerate version: 0.24.1
• xFormers version: not installed
• Using GPU in script?:
• Using distributed or parallel set-up in script?: `

Who can help?

@DN6 @yiyixuxu @sayakpaul @patrickvonplaten

[kremHabashy]

Similar error when importing anything from diffusers.models:

---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
[/usr/local/lib/python3.10/dist-packages/diffusers/utils/import_utils.py](https://localhost:8080/#) in _get_module(self, module_name)
    709         try:
--> 710             return importlib.import_module("." + module_name, self.__name__)
    711         except Exception as e:

13 frames
NameError: name 'PreTrainedModel' is not defined

The above exception was the direct cause of the following exception:

RuntimeError                              Traceback (most recent call last)
[/usr/local/lib/python3.10/dist-packages/diffusers/utils/import_utils.py](https://localhost:8080/#) in _get_module(self, module_name)
    710             return importlib.import_module("." + module_name, self.__name__)
    711         except Exception as e:
--> 712             raise RuntimeError(
    713                 f"Failed to import {self.__name__}.{module_name} because of the following error (look up to see its"
    714                 f" traceback):\n{e}"

RuntimeError: Failed to import diffusers.models.autoencoder_kl because of the following error (look up to see its traceback):
name 'PreTrainedModel' is not defined

[yiyixuxu]

ohh sorry I'm not able to reproduce this error

cc @sayakpaul maybe

[giacomov]

Same here. Can't import any model on the latest release. Gonna try reverting back to older releases, but something like this should have probably been caught by the tests :-)

[giacomov]

FYI, the problem is NOT present in version 0.21.4, so reverting to that fixes the problem for now.

[sayakpaul]

Installing transformers should fix the issue as of now. But I will investigate more deeply.

Cc: @DN6

[sayakpaul]

This is indeed quite weird. We have a dedicated workflow just for checking if the pipelines can be imported correctly: https://github.com/huggingface/diffusers/blob/main/.github/workflows/pr_dependency_test.yml

It actually errors out, too, when using the latest diffusers release:
https://colab.research.google.com/gist/sayakpaul/8418c8d7a17281ab393cd76e9632bdc2/scratchpad.ipynb

But ... on my Mac, locally, it doesn't error out. I, of course, tested it in a fresh environment with diffusers installed like so: pip install -e . from the main branch. It doesn't error out either on v0.22.0 branch (diffusers was installed using the above fashion, no transformers, and accelerate) 🤯

For now, I have opened up #5668, and here's where you can find a test that it is working:
https://colab.research.google.com/gist/sayakpaul/d1d8f8d22a024506e33515317dc10a43/scratchpad.ipynb

@giacomov @WiNE-iNEFF @kremHabashy could y'all please check on your ends too? Many thanks!

[wmthor]

Installing transformers didn't work for me. But it's quite strange, installing 0.21.4 fixes it, even when changing back to the latest version 0.22.1 after that.

Running on Google Collab environment.

[sayakpaul]

Does installing from #5668 branch resolve it? If not, could you provide a reproducible Colab?