chiehpower/Setup-deeplearning-tools

JetPack 4.6 - Python 3.8 - TensorRT works or not?

chiehpower opened this issue · 2 comments

Goal

Use Python3.8 to implement TensorRT on Jetson Device

Describe

According to JetPack v4.6, as we can see, it could support TensorRT v8.0.1 already.

2021-12-0355

On Amd64, we can know that TensorRTv8.0.1 can work on Python3.8. Ref

Unfortunately, the JetPack v4.6 was still using Python3.6; hence, I manually installed Python3.8 to try.

Problem

I tried to copy TensorRT from Python3.6, but obviously it cannot work because of complied issue.

2021-12-0309

Env

  • Device: AGX Xavier
  • JetPack v4.6
  • Python 3.6 / 3.8
  • TensorRT 8.0.1

Experiment - 1

I tried to manually install TensorRT by Python3.8, but we could see that it is for CUDA 11.3 (JetPack v4.6 is CUDA10.2)

2021-12-0358

Install TensorRT by Python3.8

2021-12-0303

Try to convert the tensorrt model.

2021-12-0311

Failure...

AFAIK, Only TensorRT 8 is able to support Python3.8. (Info from AMD64)