Not an expert in machine learning, but AFAIK diffusion models use a completely different architecture, therefore you can't use the same code to run optimized ve

Not an expert in machine learning, but AFAIK diffusion models use a completely different architecture, therefore you can't use the same code to run optimized versions of both. But maybe the core ideas can be adapted to diffusion somehow.