[Issue]: ERROR Exception: .numpy() is not supported for tensor subclasses. about automatic HOT 4 CLOSED

@Disty0 any idea about this one?

from automatic.

I de-checked the option "OpenVINO disable model caching" because it is known to produce errors in the second run. (https://community.intel.com/t5/Intel-Distribution-of-OpenVINO/OpenVINO-cannot-run-on-gpu-device-stable-diffusion-Script/m-p/1551978)

That option is already disabled by default. Can you try enabling it? Enabling it will disable caches, you did the opposite of what the linked issue page says.

You can also try unchecking VAE from compile models.

SDXL is pretty messy with OpenVINO.

from automatic.

<<That option is already disabled by default.
Ah ok, I enabled it, right after installing sd.next.

What i just experienced is, the error disappears if i switched caching on. Than i switched the cache off, restarted the server and in the second image generating the error appears again. Than i enabled the caching option again, and now it works again, no error in the second generation. Very strange, because yes, thats the opposite as in the link!
Anyhow, if you have the cache enabled by default, than maybe it s good to let it that way.

<<You can also try unchecking VAE from compile models.
ah ok, big thanks!!!!

from automatic.

seems this can be closed?

from automatic.