⚠️ This repo has been deprecated. Please use the Deep Learning Ultra container instead.

For updated code goto: https://github.com/salinaaaaaa/Deep-Learning-Ultra

• There are working notebook examples on how to wire up, both Torch and TF2 to Tensorboard in /app folder.

• Miniconda: Accelerated Python, version 3.11
• CuPy: GPU accelerated drop in replacement for Numpy
• OpenCV, latest version which is made to compile for CUDA GPUs in the container. Depending upon your GPU you may have to change -DCUDA_ARCH_BIN=7.5 in the OpenCV flags within the Dockerfile, and rebuild the image.
• PyTorch 2.0 with Torchvision for GPU, latest version
• PyTorch geometric for GNN's
• Captum to explain Torch models
• Tensorflow 2 with Keras
• Tensorboard for both Torch and TF2
• NVIDIA TensorRT inference accelerator for Tensor core access and CUDA 11 for GPUs
• Repo includes two working notebook examples on how to wire up Torch and TF2 to TensorBoard, located in /app folder

Press tab to see what methods you have access to by clicking tab.

Link to nvidia-docker2 install: Tutorial

You must install nvidia-docker2 and all it's deps first, assuming that is done, run:

sudo apt-get install nvidia-docker2

sudo pkill -SIGHUP dockerd

sudo systemctl daemon-reload

sudo systemctl restart docker

How to run this container:

docker build -t <container name> . < note the . after

If you get an authorized user from the docker pull cmd inside the container, try:

$ docker logout

...and then run it or pull again. As it is public repo you shouldn't need to login.

Run the image, mount the volumes for Jupyter and app folder for your fav IDE, and finally the expose ports 8888 for Jupyter Notebook:

docker run --rm -it --gpus all --user $(id -u):$(id -g) --group-add container_user --group-add sudo -v "${PWD}:/app" -p 8888:8888 -p 6006:6006 <container name>

:P If on Windows 10:

winpty docker run --rm -it --gpus all -v "/c/path/to/your/directory:/app" -p 8888:8888 -p 6006:6006 <container name>

Disclaimer: You should be able to utilize the runtime argument on Docker 19+ as long as it is installed and configured in the daemon configuration file:

Install nvidia-docker2 package https://github.com/nvidia/nvidia-docker/wiki/Installation-(version-2.0)#ubuntu-distributions-1

• Open another ssh tab, and exec into the container and check if your GPU is registering in the container and CUDA is working:

• Get the container id:

docker ps

• Exec into container:

docker exec -u root -t -i <container id> /bin/bash

• Check if NVIDIA GPU DRIVERS have container access:

nvidia-smi

• Check if CUDA is working:

nvcc -V

• Exec into the container as stated above, and run the following:

tensorboard --logdir=//app --bind_all

• You will recieve output that looks somnething like this:

TensorBoard 2.1.0 at http://af5d7fc520cb:6006/

Just replace af5d7fc520cb with the word localhost and launch in the browser, then you will see:

AppArmor on Ubuntu has sec issues, so remove docker from it on your local box, (it does not hurt security on your computer):

sudo aa-remove-unknown

Install the the nvidia-conatiner-runtime package, install and set-up config is here: https://github.com/NVIDIA/nvidia-container-runtime.

sudo apt-get install nvidia-container-runtime

sudo vim /etc/docker/daemon.json

Then , in this daemon.json file, add this content:

{
  "default-runtime": "nvidia"
  "runtimes": {
    "nvidia": {
      "path": "/usr/bin/nvidia-container-runtime",
       "runtimeArgs": []
     }
  }
}

sudo systemctl daemon-reload

sudo systemctl restart docker

Method 2: Install the container runtime: https://github.com/NVIDIA/nvidia-container-runtime#ubuntu-distributions

Modify the config file: https://github.com/NVIDIA/nvidia-container-runtime#daemon-configuration-file