Unary Layer in TensorRT 3.0 RC

Autonomous Machines Jetson & Embedded Systems Jetson TX2
21

Hie,

@AastaLL

I am currently using tensorflow-gpu==1.11.0 version
CUDA 9.0, TensorRT :4.0.1.6 , CUDNN :7.3.1.

Is this due to version problem??Or due to our architecture model??
What do you mean by TX2??
We are currently running on Quadro P4000 and TitanV. On both these GPU’s ,same error is getting raised.

Suggest some modifications or required to install new version.

22

Hi,

This issue is fixed in TensorRT5.0 GA.
We have different package for desktop GPU and Jetson.

You can upgrade the TensorRT x86 packaged into TensorRT5.0 GA to have the fix for this issue.
But please wait for our next JetPack release to get the TensorRT5.0 GA for ARM platform.

Thanks.

23

According to the docs for TensorRT 5.0.2 (desktop) and 5.0.3 (auto/L4T) this is still not implemented (see supported ops, tensorflow, note about NvUffParser [url]https://docs.nvidia.com/deeplearning/sdk/tensorrt-archived/tensorrt-503/tensorrt-developer-guide/index.html#support_op[/url]). Are these unary operations implemented for non-constants or not?

24

Hi,

Sorry that there is some mis-understanding in the previous comment.

There are two kind of ‘fix’ here:

1. We update the limitation of unary layer in our document.
This is fixed in TensorRT 5.0 GA. You can find this information is already listed here:
[url]https://docs.nvidia.com/deeplearning/sdk/tensorrt-archived/tensorrt-503/tensorrt-developer-guide/index.html#support_op[/url]

2. Unary for non-constant node support.
This is implemented in our internal branch.
It will be available on the next TensorRT release.

Thanks.