Diffusers Roadmap?
Hello @lodestones ,
I was just wondering how progress is going on enabling diffusers inference?
Tools like LocalAI, InvokeAI, utilize this more portable back-end, and was just wondering how progress is going.
I.E: what has been achieved so far? what is left to do?
Also if there are any issues you want me to work on to aid in this goal i am available to help, although my experience with Diffusion Models specifically is limited. I have a 40xx GPU at the moment, i know that cant be used for training dense models like this, but i hope i could be of some use.
P.S: Thankyou @lodestones i would like to extend my gratitude for your amazing work in this area, if you need anything from me, let me know.
Many thanks,
James Clarke
@Impulse2000 i think custom diffusion pipeline is good? i'm still preoccupied on tinkering with data and training rn.
but in general inference code is pretty simple
https://github.com/lodestone-rock/flow/blob/master/src/trainer/train_chroma.py#L320-L426
@Impulse2000 Here's some working but unoptimized diffusers inference code as a starting point:
https://gist.github.com/josephrocca/385d9868ac52ea6f854b3ab96ec0ad25
Example output from running that script as-is:
An aesthetically pleasing digital painting of a cat, holding a sign that says "Chroma". It has a charmingly painterly style, with visible brush strokes.
