Installation

3.1

Using Clara requires the following:

Operating system

Clara Train requires Linux having been designed on Ubuntu, and Windows is not a supported platform for Clara Train.

Driver requirements

Clara is based on NVIDIA CUDA 11.0.3, which requires NVIDIA Driver release 450. However, if you are running on Tesla (for example, T4 or any other Tesla board), you may use NVIDIA driver release 418.40.04+ or 440.33.01+. Compatibility may not be supported with certain later versions. To see details on compatibility, see CUDA Compatibility.

Note that upgrading your driver may require you to upgrade docker. Upgrading docker can be done by following the readme of this page (https://github.com/NVIDIA/nvidia-docker). If you are using a DGX system, you can follow this: https://docs.nvidia.com/deeplearning/frameworks/preparing-containers/index.html.

For more information, see NVIDIA container image of TensorFlow, release 20.08.

GPU requirements

Clara supports CUDA compute capability 6.0 and higher. This corresponds to GPUs in the Pascal, Volta, Turing, and Ampere families. For a list of GPUs that corresponds to this compute capability, see CUDA GPUs. For additional support details, see Deep Learning Frameworks Support Matrix.

Download the docker container using these commands:

Copy
Copied!
            

export dockerImage=nvcr.io/nvidia/clara-train-sdk:v3.1.01

Copy
Copied!
            

docker pull $dockerImage

Once downloaded, run the docker using this command:

Copy
Copied!
            

docker run -it --rm --shm-size=1G --ulimit memlock=-1 --ulimit stack=67108864 --ipc=host --net=host --mount type=bind,source=/your/dataset/location,target=/workspace/data $dockerImage /bin/bash

The docker, by default, starts in the /opt/nvidia folder. To access local directories from within the docker, they have to be mounted in the docker.

To mount a directory, use the -v <source_dir>:<mount_dir> option. Here is an example:

Copy
Copied!
            

docker run --shm-size=1G --ulimit memlock=-1 --ulimit stack=67108864 -it --rm -v /home/<username>/clara-experiments:/workspace/clara-experiments $dockerImage /bin/bash

This mounts the /home/<username>/clara-experiments directory in your disk to /workspace/clara-experiments in docker.

Hint

More information for mounting directories can be found in Docker documentation

If you are on a network that uses a proxy server to connect to the Internet, you can provide proxy server details when launching the container.

Copy
Copied!
            

docker run -it --rm -e HTTPS_PROXY=https_proxy_server_ip:https_proxy_server_port -e HTTP_PROXY=http_proxy_server_ip:http_proxy_server_port --shm-size=1G --ulimit memlock=-1 --ulimit stack=67108864 $dockerImage /bin/bash

For GPU isolation in the docker, use --gpus= with the latest docker release as shown here.

Copy
Copied!
            

docker run -it --rm --gpus=1 --shm-size=1G --ulimit memlock=-1 --ulimit stack=67108864 $dockerImage /bin/bash

The NGC models page has models available for direct download as a zip archive.

You can also download models from inside the container with the built-in NGC commands to retrieve or list models hosted on NGC.

Use this command to list available models: ngc registry model list nvidia/med/*

Copy
Copied!
            

root@03c5db1ddbcc:/opt/nvidia# ngc registry model list nvidia/med/*

Choose a model and set the MODEL_NAME and VERSION for that model, then to download the model to the current directory, use the command below:

Copy
Copied!
            

MODEL_NAME=clara_mri_seg_brain_tumors_br16_full_amp VERSION=1 ngc registry model download-version nvidia/med/$MODEL_NAME:$VERSION --dest /var/tmp Downloaded 49.74 MB in 4s, Download speed: 12.4 MB/s ---------------------------------------------------- Transfer id: clara_mri_seg_brain_tumors_br16_full_amp_v1 Download status: Completed. Downloaded local path: /var/tmp/clara_mri_seg_brain_tumors_br16_full_amp_v1 Total files downloaded: 22 Total downloaded size: 49.74 MB Started at: 2020-01-13 19:01:06.519897 Completed at: 2020-01-13 19:01:10.526815 Duration taken: 4s ----------------------------------------------------

Browse the NGC models page for the latest model updates and information.

© Copyright 2020, NVIDIA. Last updated on Feb 2, 2023.