OSError: Unable to load weights from pytorch checkpoint file for (...)

[furmanlukasz]

In direct linux machine I get these error:

OSError: Unable to load weights from pytorch checkpoint file for '/home/huggingface/.cache/huggingface/diffusers/models--CompVis--stable-diffusion-v1-4/snapshots/48a1bd8a2d86483de1ac27e95e188569b57a494e/unet/diffusion_pytorch_model.bin' at '/home/huggingface/.cache/huggingface/diffusers/models--CompVis--stable-diffusion-v1-4/snapshots/48a1bd8a2d86483de1ac27e95e188569b57a494e/unet/diffusion_pytorch_model.bin'. If you tried to load a PyTorch model from a TF 2.0 checkpoint, please set from_tf=True.

Looks like this path is broken when building it /home/huggingface/.cache/huggingface/diffusers/models--CompVis--stable-diffusion-v1- but I can pind place where to fix it.

[fboulnois]

To clarify, are you trying to run the Python code without Docker or the build script?

[furmanlukasz]

This is after building docker container when running ./build run "prompt"

[fboulnois]

I reran the commands on a fresh machine and I wasn't able to reproduce this issue. If you clear out the huggingface volume (docker volume rm huggingface), are you able to download all the models successfully to fix the issue? Keep in mind this will permanently erase the volume and all its contents.

[furmanlukasz]

That did work! Thank you, however right now it seems that docker has no permission to save it in output directory PermissionError: [Errno 13] Permission denied: 'output/horse_on_the_moon___steps_50__scale_7.50__seed_14724727502852850856__n_1.png'

[fboulnois]

Were you able to resolve this issue? Does the output folder exist outside the container? If so, which user owns that folder?

[furmanlukasz]

I need to add the docker to a new group, then it had permission to write down the file.