Google’s Deep Learning Infrastructure Stack with Facebook’s Pytorch

Develop Your Own PyTorch App

in the Cloud with Free GPUs

Step 1: Install the Google Colaboratory app in chrome browser

This step is as simple as launching your chrome browser and clicking this link to install chrome extension. This chrome extension is based on jupyter notebook, an open source web development environment for machine learning. Click on ‘ADD TO CHROME’ will install the development environment in your chrome browser.

Install Google Colab App in Chrome

Step 2: Create/ Open the Colaboratory Notebook Following URL contains Colaboratory notebook with open source code for estimating capacitance based on physical characteristics of a wire. Once you click on the link https://pytorch.org/tutorials/beginner/blitz/tensor_tutorial.html#getting-started, it opens up in your chrome browser as shown in the picture.

PyTorch Tutorial (credit: pytorch.org)

Step 3: Modify app to add your functionality.

Take this notebook, delete the code and start adding your custom functionality. I successfully used object detection using yolo algorithm using pytorch in colab.

yolo object detection using pytorch in google colab

Step 4: Use GPUs for free

If your app is taking longer than usual, you can use GPUs for free by hitting “Runtime -> Change runtime type” as shown in the picture below:

Using GPUs in Google Cloud

Summary

Now you’ve developed your first pytorch app in the cloud with GPU provided by google, consider yourself armed with all the information needed to deploy deep learning in your next application. Good luck !

Reference:

1. PyTorch: A Deep Learning Research Platform
2. Book on Machine Intelligence in Design Automation
3. Short Course on Machine Intelligence in EDA and CAD
4. Open source machine learning apps for EDA and CAD