I cannnot load libcudart.so.10.2 to Docker container based on l4t-tensorflow

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
,
1

Hi, I am a beginner with Jetson devices. I know this is an older device, but I am using a Jetson Nano 4GB and trying to set up a container based on l4t-tensorflow.

I want to use TensorFlow and TensorRT inside the container, but when I try to import tensorflow, I get the following error:

libcudart.so.10.2: cannot open shared object file: No such file or directory

Could you please tell me how to properly load libcudart.so.10.2 inside the Docker container?

Environment:

  • JetPack 4.6.1 (L4T R32.7.1)
  • Docker Image: l4t-tensorflow:r32.7.1-tf2.7-py3
  • Device: Jetson Nano 4GB
  • Command: sudo docker run -it --rm --runtime=nvidia --gpus all my-tf-container:0.1

What I Have Tried:

  1. The host system has the correct libcudart.so.10.2. But it is not mounted on container(in spite of that, libcudart_static.a is mounted successfully)
  2. I set the library path (LD_LIBRARY_PATH), but it is still not recognized.
  3. I copied the libcudart.so.10.2 file from the host to the container, but it is still not recognized.
3

Hi,

It should mount into the container when running with nvidia runtime.
Does the sudo docker run -it --rm --runtime=nvidia --gpus all my-tf-container:0.1 the command you use?

Could you try it with the following command:

$ sudo docker run -it --rm --runtime nvidia --network nvcr.io/nvidia/l4t-tensorflow:r32.7.1-tf2.7-py3

Thanks.

4

Thank you for your reply,

I execute the command you suggested, but It’s still not resolved.

sudo docker run -it --rm --runtime nvidia --network host nvcr.io/nvidia/l4t-tensorflow:r32.7.1-tf2.7-py3

result:
In the container (created without warnings), libcudart.so.10.2 is still missing.

find / -name  "*libcudart*"
/usr/local/cuda-10.2/targets/aarch64-linux/lib/libcudart_static.a
5

I solved this problem.
After checking my environment, I found that I was actually using JetPack 4.6.6 (L4T R32.7.6), which was incompatible with the l4t-tensorflow version.
So, I installed JetPack 4.6 (L4T R32.6.1) on another SD card, used l4t-tensorflow:r32.6.1, and it worked.

6

Hi,

Thanks for the feedback.

We also have packages for JetPack 4.6.1 (r32.7.1) in the below link:

Thanks.

7

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.