NPU Support?
SirHenryAvery opened this issue · 0 comments
Is your feature request related to a problem? Please describe.
I don't have a problem, other than my own user errors. I have a request: Would it be possible to run a modified (or native?) version of easy-diffusion on a Khadas single board computer, or similar NPU/TPU? If not, what are the hurdles? Are there other similar devices that do support ED? If so, what are they? What are the potential benefits or shortcomings?. If I could contribute cash to assist, how much are we talking about, and how would financial support help?
Describe the solution you'd like
To be able to run ED efficiency on a Khadas SBC, or on a networked series of such units.
Describe alternatives you've considered
Nothing. I'm an ordinary dude that doesn't know how to code, and python is utterly beyond my understanding and capabilities.
Additional context
I'm willing to put my money where my mouth is. I can provide test hardware for any dev that wants to venture into this particular space of Never Never Land.
Oh, if these devices are not "neural processing units" or "neural network processing units", or are otherwise utterly unsuitable for use in this case scenario, please publicly ridicule me - I probably deserve it.
These are the specific devices I am considering: