Inference speed

[tppqt]

I've used the Open-CLIP model for sports competition ReID, but I'm looking to improve the inference speed to achieve real-time performance, as it currently takes about one second per data run. Are there any methods?

[rwightman]

@tppqt the usual things will likely help, compile, tensorrt, quantization ... but would need feedback from other users who have tried them.