I have an AGX Xavier running the latest JetPack - I think it’s 4.5.1; I just re-flashed it last week. Which, if any, JetPack software packages can use the NVDLA engines? Do they work with PyTorch, for example?
Only TensorRT support DLA inference.
Please convert the PyTorch model into TensorRT compatible first.
A common use workflow is:
- Convert .pth file into .onnx with the instruction shared here.
- Generate the TensorRT engine for DLA with trtexec as below:
$ /usr/src/tensorrt/bin/trtexec --onnx=/usr/src/tensorrt/data/mnist/mnist.onnx --useDLACore=0 --allowGPUFallback
Thanks!! I sort of figured the ONNX part out from the documentation; I’m doing audio processing and I haven’t decided whether to use TensorFlow or PyTorch. The end goal is something like this:
ONNX is the Open Neural Network Exchange (ONNX).
It allows different frameworks to sharing the model.
Almost all the frameworks can convert their default file into ONNX.
For example. tf2onnx for TensorFlow model.