OSError: CompVis/stable-diffusion-v1-4 does not appear to have a file named config.json.

[XavierXiao]

When using SD Diffusers, I got errors when running this line

text_encoder = CLIPTextModel.from_pretrained( 'CompVis/stable-diffusion-v1-4', subfolder="text_encoder", use_auth_token=True ),

where the error is

OSError: CompVis/stable-diffusion-v1-4 does not appear to have a file named config.json.

Any idea on why this happens?

[apsifly]

Faced this issue myself. Your version of transformers package is probably old and does not recognise subfolder parameter.
Not related to this repo though.

[successbyfailure]

+1
Following the guide https://github.com/huggingface/diffusers/tree/main/examples/textual_inversion I also run into that error.

Im using transformers from pip... so.... its waiting till they release a new version?

[apsifly]

already fixed at least in 4.22.1

[successbyfailure]

You're right!
The conda env has transformers version pinned to 4.19.2. Updated and solved :)

[gregedout]

so what should we do if we face this error?

[successbyfailure]

Say not today and upgrade transformers.

For me the case was that I had the environment already setup and transformers was pinned to 4.19 in the requirements.txt

So I just edited the requirements.txt and set transformers to at least 4.22.1 then run pip install -r requirements.txt, that did the trick for me

[sjkoelle]

fwiw im getting this error with transformers 4.27.1

[squewel]

Also having issues on the latest version

[MrHup]

Still happening as of 27 May with latest transformers version.

[wyf-gh]

Also having the same issues on transformers==4.36.0 cuda==11.6 pytorch==1.12.0. Does anyone have a solution?

[VikingMew]

meet in 4.37.1

[Value-Jack]

meet in 4.37.1

meet too! 4.37.1 do you solve this problem?

[DetoitBecomeHuman]

me too in 4.31.

[mlmlh]

same for 4.39.1

[BingliangLi]

This can be internet issues, the below solved for me, run this in command line:
HF_ENDPOINT=https://hf-mirror.com python # Or just add this before your xx.py and ignore the rest.

Then run the code to cache the model(change the model card name to what you need)
text_encoder = CLIPTextModel.from_pretrained( 'CompVis/stable-diffusion-v1-4', subfolder="text_encoder", use_auth_token=True )
Now run you code.
Env:

diffusers == 0.27.1
accelerate == 0.28.0
transformers == 4.35.2

[Forence1999]

I met this error when I use "stabilityai/stablelm-2-1_6b", because I build a folder with the path "stabilityai/stablelm-2-1_6b", which will guide transformers package to load model from the local folder. After removing the folder, the error disappears.

[Jeon-jisu]

In my case, upgrading the transformers version did not resolve the issue, so I focused on the 'subfolder="unet"' parameter.
Originally, the code was written as :
UNet2DConditionModel.from_pretrained(config["stable_diffusion_path"])
I modified it to:
UNet2DConditionModel.from_pretrained(config["stable_diffusion_path"], subfolder="unet")
After this change, I cleared the Hugging Face cache using the following command:
rm -rf ~/.cache/huggingface/hub
Then, I ran the original file I was trying to execute. This approach proved effective in resolving the issue.

[adityapandey2811]

For me, I updated:

UNet2DConditionModel.from_pretrained(model_id)
to
UNet2DConditionModel.from_pretrained(model_id, subfolder="unet")

and

DDPMScheduler.from_pretrained(model_id)
to
DDPMScheduler.from_pretrained(model_id, subfolder="scheduler")

This worked for me.