Google Compute Engine Now Has Images With PyTorch 1.0.0 and FastAi 1.0.2

Today we are announcing a huge update to our image families(if you do not know what Deep Learning images are you might want to read the following article first). New experimental images with the following framework pre-installed:

• PyTorch 1.0.0 Preview and
• FastAi 1.0.2

But this is not all, images are also come with pre-installed tutorials for both PyTorch and FastAi. For example, here is vision notebook, in the fastai folder, running on the Jupyter Lab (on CPU instance):

There are 2 images, one for CPU and one for GPU with CUDA pre-installed. Image families are:

* pytorch-latest-gpu
* pytorch-latest-cpu

As usual, there are two ways to use the image.

Start Via UI (Google Cloud Marketplace)

Go to the Marketplace page of Deep Learning images. Click “launch on compute engine”:

Now set the framework to PyTorch 1.0

Fill up the rest and click “Deploy”. This is it, now you have a Deep Learning VM with PyTorch 1.0 and FastAi 1.0 pre-installed.

Start Via Cli

If you do not have yet, install gcloud cli. Now, if yo need GPU instance use the following command:

For the CPU instance you can use this:

If you want to learn more about how to create deep learning VM from the shell, read this article.

Hot To Use Jupyter Lab

First you need to install gcloud cli. There is a way to use Jupyter Lab without installing gcloud locally, it can be found here.

As soon as instance created you can SSH to it:

gcloud compute ssh $INSTANCE_NAME -- -L 8080:localhost:8080

and open your browser at http://localhost:8080 . This is!