[Feature Request]:

[Metsuryu]

Is there an existing issue for this?

• [X] I have searched the existing issues and checked the recent builds/commits

What would your feature do ?

There is a recent paper that presents a series of implementation optimizations for large diffusion models that achieve the fastest reported inference latency to-date:

https://arxiv.org/abs/2304.11267

It would be great to implement these optimizations.

Proposed workflow

It should work by default, after implementing the optimizations described in the paper.

Additional information

No response

[BinaryQuantumSoul]

You should also ask for vladmantic https://github.com/vladmandic/automatic

[Metsuryu]

I did

[anapnoe]

cool I wish there was an actual implementation on this on how compute shaders are utilized