How to auto tune a convolutional network for nvidia gpus

Questions
1

How to auto tune a convolutional network for nvidia gpus?

2

I am cleaning up the code.
I think I can send the code, benchmark and tutorial in two weeks.

3

I’ll push PR for graph level tuner in one or two weeks. It can be combined with autotvm to get optimal performance. It would be great if we can benchmark TVM vs TensorRT for NVIDIA gpu.

4

@tianxingyzxq The update (https://github.com/dmlc/tvm/pull/1638) is merged with tutorial https://docs.tvm.ai/tutorials/autotvm/tune_nnvm_cuda.html

5

Thank you for your work!
As you said,you benchmark models using MXNet +Tensor-rt 4.0
However,I want to know tensorrt takes FP32 or INT8?