unet\mv_unet.py as defined in model_index.json does not exist in ashawkey/mvdream-sd2.1-diffusers

[ackmonra]

Hi there.

I'm getting this error trying to boot the app. I'm going to guess I've either got a wrong version; or missed something - but when it goes to load this pretrained:

load dreams

pipe_text = MVDreamPipeline.from_pretrained(
'ashawkey/mvdream-sd2.1-diffusers', # remote weights
torch_dtype=torch.float16,
trust_remote_code=True,
# local_files_only=True,
)

in app.py

I end up with:

ValueError: unet\mv_unet.py as defined in model_index.json does not exist in ashawkey/mvdream-sd2.1-diffusers and is not a module in 'diffusers/pipelines'.

Any ideas where I'm going wrong? I checked and the hugging face entries look ok.

Thanks.

[ashawkey]

@ackmonra Hi, could you update diffusers and try again?

[ackmonra]

It could be a problem for me - I already use 2.1.1+cu121 with xformers-0.0.22 for other models and they have different xformers dependancy. It might be hard for me to change.
I had some success directly downloading the hugging face files and specifying the models on my hard drive when doing from_pretrained.
So I think my problem is solved for now that way. It appears to work; though I am not getting the quality I was hoping for from the result; perhaps that is an artefact of using the unintended torch versions etc.

[ashawkey]

I guess torch version may not affect the final quality. You can compare with the gradio demo for verification.