Thanks for your work.
I tried to train FontDiffuser and use its web UI, and ran into some problems doing so. These are changes I find needed in requirements.txt:
- Add fastapi==0.109.2
When installing gradio==4.8.0, pip/conda will by default install fastapi 0.115.*. This version causes the error Exception in ASGI application AttributeError: __pydantic_core_schema__ when interacting with the examples or outputting the generated image.
Specifying fastapi 0.109.2, as suggested by this post, fixes the problem.
- Add tensorflow<=2.17.1
This is required for the use of tensorboard to view training log during or after training.
- Add huggingface_hub<0.26.0
Additionally, some imports will not work without restricting the version of huggingface_hub due to newer releases.
- Use kornia<=0.7.3 or a later version.
Do not use kornia 0.7.4 or 0.8.0. There was a problem with the library interfering with logging, causing the loss of some logs that should otherwise go to output_dir/fontdiffuser_training.log.
I hope this can serve well as a bug report/tips for future users.
Thanks for your work.
I tried to train FontDiffuser and use its web UI, and ran into some problems doing so. These are changes I find needed in
requirements.txt:When installing gradio==4.8.0, pip/conda will by default install fastapi 0.115.*. This version causes the error
Exception in ASGI applicationAttributeError: __pydantic_core_schema__when interacting with the examples or outputting the generated image.Specifying fastapi 0.109.2, as suggested by this post, fixes the problem.
This is required for the use of tensorboard to view training log during or after training.
Additionally, some imports will not work without restricting the version of huggingface_hub due to newer releases.
Do not use kornia 0.7.4 or 0.8.0. There was a problem with the library interfering with logging, causing the loss of some logs that should otherwise go to
output_dir/fontdiffuser_training.log.I hope this can serve well as a bug report/tips for future users.