How to run ML/DL model on remote GPU server in Jupyter notebook

If your AWS, Azure or any other remote server has GPU, you would like to run you model on them.

Maybe tensorflow, keras, or sklean.

Just follow these steps:

Step 1

Install Docker and Cuda driver on your server.

Step 2

Run tensorflow and jupyter container on your server.

docker run -p 8888:8888 quay.io/jupyter/scipy-notebook:2023-10-31

For more detail on this Docker image, please refer to docker-stacks.

Step 3

From previous step, you will get the output from docker run, it contains the token:

http://127.0.0.1:8888/lab?token=aa821e54884537ec41eb845c0cfaa5369332dc02c59f2b59

replace 127.0.0.1 to public IP of the remote server.

Open it on browser, you will see a remote jupyter.

Step 4

Upload Jupyter notebook. Upload the .ipynb file using browser and run it.

If you want to install denpendencies, just use !pip.

!pip install --upgrade pip
!pip install numpy tensorflow
!pip install imageio scikit-image