Closed vvolhejn closed 2 years ago
Probably just on one runtime, ideally the best one
For ONNX Runtime:
The conclusion is that if the models' runtime is at least 1 ms, the overhead is small (<10%).
Probably just on one runtime, ideally the best one