Is there an existing issue for this?
Contact Details
No response
What should this feature add?
Add support for Pytorch 2.0 nightly, and possibly its new features like torch.compile() and the inductor (and/or various other) optimized backends.
Its early, yes, but it would be nice to have the performance and compatibility improvements now, and InvokeAI compatibility (at least partially) ready to go when Pytorch 2.0 is officially released.
Currently it doesn't work with Invoke AI for me, but it does work with a vanilla Diffusers StableDiffusionImg2ImgPipeline. See the odd error I ran into here: #2396
See also the issues on the diffusers side, as well as a long discussion from Automatic's repo:
huggingface/diffusers#1964
AUTOMATIC1111/stable-diffusion-webui#6932
I would also be interested to see if the current nightly build works for anyone, or if I am encountering some bizzare environment issue.
Is there an existing issue for this?
Contact Details
No response
What should this feature add?
Add support for Pytorch 2.0 nightly, and possibly its new features like torch.compile() and the inductor (and/or various other) optimized backends.
Its early, yes, but it would be nice to have the performance and compatibility improvements now, and InvokeAI compatibility (at least partially) ready to go when Pytorch 2.0 is officially released.
Currently it doesn't work with Invoke AI for me, but it does work with a vanilla Diffusers StableDiffusionImg2ImgPipeline. See the odd error I ran into here: #2396
See also the issues on the diffusers side, as well as a long discussion from Automatic's repo:
huggingface/diffusers#1964
AUTOMATIC1111/stable-diffusion-webui#6932
I would also be interested to see if the current nightly build works for anyone, or if I am encountering some bizzare environment issue.