why i use A100 80G to inference so slow?

[Arbor334]

I downloaded all the files and data locally, but the graphics card memory did not increase when loaded, and the reasoning was very slow, especially when loading initialize_model functions, and now I don't have a complete single_video_inference.py file