Support for pixart-ai under 8Gb of VRAM

[MOULINIER]

Hi, could it be possible to implement the method used in the pixart GitHub to make the model able to work under 8Gb of VRAM ?
I could try it but I'm scared to break things ( I'm not really experienced with python), would the push request be accepted?

[vladmandic]

PRs are more than welcome - I will review anything that is submitted and if it helps, it will be merged.