I use apps/benchmark to do some test. I try to replace tvm.relay.testing with tvm.relay.frontend.from_tflite, then I see a growth of time when evaluating the same model.
For example, resnet-v2-50, time grows from 165ms to more than 4000ms.
What’s more, when I use relay.frontend.from_tflite, benchmark log shows that:
“One or more operators have not been tuned. Please tune your model for better performance.”
So I am wondering what is the difference between them?