onnxruntime

[CUDA] stable diffusion benchmark allows IO binding for optimum

#22834

Merged