Your video card does not support half type

[goshante]

I have RTX 2070 Super. When I am using most of models it's okay, but some models cause errors.
For example AbyssOrangeMix3.
When i generate simple image with thos model without any lora, any overcomplication, etc, i got this error:
modules.devices.NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

So what I have now:

1. --no-half does not help, i just getting shorter error "A tensor with all NaNs was produced in Unet. Use --disable-nan-check commandline argument to disable this check."
2. "Upcast cross attention layer to float32" has no effect at all
3. --disable-nan-check removes this error, but i got black rectangles instead of images.

How to make this model work for me?
The goal is just to make this model make images.

[Sakura-Luna]

I haven't encountered a Unet error so I just have a little suggestion:

1. It should only affect specific configurations, so you can try to modify parameters or configurations to get other results.
2. You can post your startup parameters, venv environment information and a specific example parameter to help others reproduce this problem.

[goshante]

@Sakura-Luna sure
Here is it: python: 3.10.9  •  torch: 1.13.1+cu117  •  xformers: 0.0.16rc425  •  gradio: 3.23.0  •  commit: 22bcc7be  •  checkpoint: 8730188de4

[Sakura-Luna]

I checked #6923 and think your problem is that the model file is corrupted, I suggest you re-download this model.

[goshante]

@Sakura-Luna OMG YOU WAS RIGHT! Model was corrupted. It works after re-downloading

[Sakura-Luna]

I think you should change the category to "Q&A". Of course, it's better to give me a mark answer :).

[linexoff]

are you suggesting reinstalling checkpoint? I have the same problem with the black rectangle and I didn't understand what "re-download this model" means