Docker image with tensorrt 10.x

Discussion
1

hi, im a newbie here, just wondering where can I find a docker image with tensorrt 10.x

I tried these 2, both still on tensorrt 8.6.2
ultralytics/ultralytics:latest-jetson-jetpack6
docker pull ultralytics/ultralytics:8.3.9-jetson-jetpack6

thanks

2

Hi there! :wave:

Welcome to the community! For Docker images with TensorRT 10.x, you might need to build a custom image. The current Ultralytics images are based on TensorRT 8.6.2. You can start by modifying the Dockerfile to include TensorRT 10.x.

Here’s a helpful guide to get you started: NVIDIA Jetson Deployment Guide.

Feel free to ask if you have more questions. Happy coding! :rocket:

3

@Gildo_Medeiros the latest Dockerfile shows it’s built from this NVIDIA NGC image) following the link referenced in the Dockerfile, there’s a statement:

The docker file for this container can be found at this link. You can refer to the [D]ockerfile and use that recipe as a reference to create your own development container (with both dev and runtime components) or deployment container (with only runtime components)

The l4t-jetpack Dockerfile shows that it installs the developer package for TensorRT which might use the older version (couldn’t find anything about this directly):

# Install nvidia-tensorrt-dev for TensorRT developer 
# packages. Use nvidia-tensorrt if need TensorRT 
# runtime only
#
RUN apt-get update && apt-get install -y \
    --no-install-recommends \
    nvidia-tensorrt-dev \
    && rm -rf /var/lib/apt/lists/* \
    && apt-get clean

You’ll either have to make some modifications after pulling from the Ultralytics image, build with the Ultralytics image after modifying the Dockerfile to remove nvidia-tensorrt-dev then installing nvidia-tensorrt, or you’ll have to create a custom, multi-stage Dockerfile to build containers from (combining the l4t-Dockerfile and Ultralytics official Dockerfile) and include the change from nvidia-tensorrt-dev to nvidia-tensorrt.

Unfortunately, I’m not terribly familiar with running things on JetPack or Jetson devices, that’s always been @lakshanthad’s wheelhouse. If you find a good way to make it work, I definitely encourage you to open a pull request on GitHub. You could also open an issue for a feature request to include a non-dev install option as well.

4

Sorry to jump in late here. @BurhanQ Thanks for the pointers here.

I am currently working on updating our JetPack6 containers to JetPack 6.1. JetPack6.1 comes with TensorRT 10.3 out-of-the-box and it should be pretty straightforward to get things going. I will do some testing, publish the updates officially and let you know here @Gildo_Medeiros.

In the meantime, Please change this line to FROM nvcr.io/nvidia/l4t-jetpack:r36.4.0 and try to build the container. This will build JetPack6.1 with TensorRT 10.3.

Thanks for the patience.

2 Likes
5

Hello @Gildo_Medeiros We have updated with JetPack6.1 which includes TensorRT 10.3 support with Docker:

This will do the job!

t=ultralytics/ultralytics:latest-jetson-jetpack6
sudo docker pull $t && sudo docker run -it --ipc=host --runtime=nvidia $t

Thanks for waiting!

CC: @BurhanQ

1 Like