The latest articles on DEV Community by IKEUCHI Yasuki (@ikeyasu). https://dev.to/ikeyasu https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F44677%2F458c5c0c-062c-4d35-a3eb-d9de11e5ec67.jpg https://dev.to/ikeyasu en IKEUCHI Yasuki Wed, 11 Apr 2018 01:37:36 +0000 https://dev.to/ikeyasu/how-to-use-google-colaboratory-with-local-runtime-4j1p https://dev.to/ikeyasu/how-to-use-google-colaboratory-with-local-runtime-4j1p <h1> TL;DR </h1> <ul> <li>I created <a href="https://hub.docker.com/r/ikeyasu/colab-local/">docker image</a> look-alike Google Colaboratory <ul> <li>It is using the same OS version(Ubuntu17.10) as Google colaboratory.</li> <li>It is installed the same python library as Google colaboratory. </li> </ul> </li> <li>You can use this docker image with “Local Runtime” feature of Google Colaboratory</li> </ul> <h3> Step to connect </h3> <ol> <li>Setup <a href="https://github.com/nvidia/nvidia-docker/wiki/Installation-(version-2.0)">nvidia-docker2</a> </li> <li> <code>$ docker run --runtime=nvidia -it --rm -p 8081:8081 --cap-add SYS_ADMIN --device /dev/fuse --security-opt apparmor=unconfined ikeyasu/colab-local:latest</code> <ul> <li> <code>--cap-add SYS_ADMIN</code>, <code>--device /dev/fuse</code>, <code>--security-opt</code> and <code>apparmor=unconfined</code> are necessary for mounting Google drive.</li> <li>You need to change port number, please change <code>-p 8081:8081</code>. For example, <code>-p 8082:8081</code> if you want to use 8082.</li> </ul> </li> <li>Click ▼ on top right of Colaboratory(next to "CONENCT" button), and click</li> <li>Input backend port to your docker instance like 8081. Then, click Connect button.</li> </ol> <h3> If a host running jupyter notebook is not localhost </h3> <p>Colaboratory requires localhost to connect local instance. If you would like to run jupyter notebook on another host, you need to use port-forwarding as follows.</p> <p><code>$ ssh user@example.com -L 8081:localhost:8081</code></p> <p>If you are using windows, you can use PuTTY.</p> <h1> What is benefit? </h1> <p>Google Colaboratory has many limitation, especially it could be ran by 12 hours only. In other hand, Google Colaboratory is good for first step of deep learning.</p> <p>Thus, you can take a following step to move forward your project.</p> <ol> <li>Use Google Colaboratory for rough investigation and testing. It has limitation of 12 hours.</li> <li>If your model is looks good, launch Local Runtime. It does not have limitations. You can run a learning process for more 12 hours.</li> </ol> <p>Docker image introduced this paper is looka-like Google Colaboratory. You don't need to concern a difference of environment between host runtime and local runtime.</p> deeplearning colaboratory