Fast.ai Stuff

Toy Web App

To run use

$ python app.py

and

$ ngrok http 5000

Running on a remote server

Running the notebook on a remote server with a GPU via another server that gives access to the GPU server. The problem is that the machine with the GPU doesn't have any ports that can be seen from the outside other than SSH. First setup port forwarding

$ ssh intermediate-machine
$ ssh -N -L intermediate-machine:8888:localhost:8888 gpu-machine

Start up the notebook

$ ssh intermediate-machine
$ ssh gpu-machine
$ cd .../fastai-nix-install
$ nix-shell --pure
[nix-shell]$ cd ...
[nix-shell]$ jupyter notebook --no-browser --NotebookApp.allow_remote_access=True

The allow_remote_access is required otherwise you get a 403. Copy the token and you should be able to view the notebooks from http://intermediate-machine:8888/?token=<token>.

Google Compute Platform

Setup: https://course.fast.ai/start_gcp.html

Note that I had to set export ZONE="us-east1-c" to get things working, but then I got the Quota 'GPUS_ALL_REGIONS' exceeded and have to wait to have that adjusted.

Start up the instance, go here

$ https://console.cloud.google.com/compute/instances?project=fastai-wd15&folder&organizationId&instancessize=50

To login into the instance

$ gcloud compute ssh --zone=$ZONE jupyter@$INSTANCE_NAME -- -L 8080:localhost:8080

This will log in and also fire up a jupyter instance which you can see from localhost:8080.

Go to

$ cd ~/tutorials/fastai/course-v3/nbs/dl1
$ jupyter notebook

View the notebooks here at http://localhost:8080/lab?