Deploy the notebook

Prerequisites

1. Nvidia Driver
2. Docker CE
3. Nvidia-Docker
4. BERT Demo Jupyter Notebook (BERT_Demo.ipynb) - downloaded from this page.

Setup

Start The environment

Given that Docker and Nvidia-Docker are setup correctly on the demo server, launching this demo notebook should be as simple as running this command:

docker run --runtime=nvidia -it --net=host -v $(pwd):/workspace \
	--name NGC-BERT-Demo-Notebook \
	nvcr.io/nvidia/tensorflow:19.07-py3 \
	jupyter notebook --ip=0.0.0.0 --allow-root --NotebookApp.token=''

This will start the Jupyter Notebook at http://: where is the IP of the server running this docker command, and will default to 8888. You should expect to see a page like below:

note - while this command will pull the 19-07-py3 version of the TensorFlow container - it may work with later versions, but the notebook is untested.

Execute the notebook

1. This will start the Jupyter Notebook at http://<IP>:<PORT> where <IP> is the IP of the server running this docker command, and <PORT> will default to 8888.
2. Now you can just click on the BERT_Demo.ipynb to open the Notebook, and just start running through the cells using either the “Run” button at the top, or by pressing <Shift>+<Enter> when a given cell is highlighted.

Authentication

You can require a password in order to login to the demo notebook by replacing --NotebookApp.token='' with --NotebookApp.token='<password>' you'll then see a password screen when you navigate to the Notebook page in your browser.