transformers being locked to 4.23.1 causing issues

[xelfer]

In setup.py, is there a reason the transformers dependency is locked to 4.23.1?

I'm hitting an error trying to build a docker worker image of Kandinsky2 with img2img support. There is a CLIPImageProcessor error which is solved by manually updating transformers to 4.29.2.

Traceback (most recent call last):
  File "//rp_handler.py", line 8, in <module>
    from diffusers import KandinskyImg2ImgPipeline, KandinskyPriorPipeline
  File "/usr/local/lib/python3.10/dist-packages/diffusers/__init__.py", line 59, in <module>
    from .pipelines import (
  File "/usr/local/lib/python3.10/dist-packages/diffusers/pipelines/__init__.py", line 45, in <module>
    from .alt_diffusion import AltDiffusionImg2ImgPipeline, AltDiffusionPipeline
  File "/usr/local/lib/python3.10/dist-packages/diffusers/pipelines/alt_diffusion/__init__.py", line 32, in <module>
    from .pipeline_alt_diffusion import AltDiffusionPipeline
  File "/usr/local/lib/python3.10/dist-packages/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion.py", line 21, in <module>
    from transformers import CLIPImageProcessor, XLMRobertaTokenizer
ImportError: cannot import name 'CLIPImageProcessor' from 'transformers' (/usr/local/lib/python3.10/dist-packages/transformers/__init__.py)

Manually updating via pip install git+https://github.com/huggingface/[email protected] fixes the issue, but gives an error which is concerning. And this does have to be done manually, it can't be done in requirements.txt.

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
kandinsky2 0.0.0 requires transformers==4.23.1, but you have transformers 4.29.2 which is incompatible.
Successfully installed transformers-4.29.2

Just wondering if the restriction can be safely removed or if the lock can be changed to 4.29.2?

[XmYx]

You might want to check my fork with mem optimization and less strict transformer, works fine on higher version.