Steps:
Install and download pretrained weights
cuda version mismatch
Solution:
export CUDA_HOME=/usr/local/cuda-12.1
or similar depending on your cuda installation
data did not match any variant of untagged enum ModelWrapper at line …
Solution:
Upgrade transformer pip install transformers==4.45.0
Suspected cause: Model file in huggingface has been updated and is no longer compatible to earlier versions of transformers. Possible alternative solution: use a history version of the model
cannot import name ‘cached_download’ from ‘huggingface_hub’
Solution:
Downgrade huggingface_hub pip install huggingface_hub==0.23.2
Reason:
Deprecation of cached_download
and
transformers 4.45.0 requires huggingface-hub<1.0,>=0.23.2
See: cannot import name ‘cached_download’ from ‘huggingface_hub’ · Issue #1851 · easydiffusion/easydiffusion Huggingface hub change log: Release v0.26.0: Multi-tokens support, conversational VLMs and quality of life improvements · huggingface/huggingface_hub
RuntimeError: NATTEN is not compiled with CUDA! Please make sure you installed correctly by referring to shi-labs.com/natten.
Solution:
goto shi-labs.com/natten , select your cuda and pytorch version to get the command for installation e.g. pip3 install natten==0.17.3+torch210cu121 -f https://shi-labs.com/natten/wheels/
Note that the version is different form what’s stated in the environment.yaml
Misc
Issues resolvable by following the error message suggestions are not listed.