Optimization of multiple models in a single container

My top two use cases so far:

  • Real ESRGAN which uses both it’s own model for upsampling and GFPGAN for face restoration.
  • CLIP based Stable Diffusion (SD model + additional CLIP model).

Of course in the interim I’ll choose the biggest model for optimization, but the unoptimized models can really slow things down. Thanks, banana-team! I know there’s no quick fix for this :sweat_smile: