py3 is fine. you have to change a few lines occasionally e.g. import cpickle=>import pickle.
The AMI can be changed to another amazon AMI in a different region - just add a copy of the config file in ~/.xdrive/config.yaml and make the changes to that. Also you can load any image you want in the docker container. Therefore there should be no need to change the AMI…if you think otherwise please tell me why?
My aim was a thin server with all the work done in the docker container. The fastai AMI is 130GB as is the amazon deep learning AMI. I tried basic ubuntu but was unable to install the GPU drivers using the 100 page installation manual! So I then tried to find an AMI with the GPU drivers installed. The only lighweight one I could find was nvidia/amazon; and despite this being nvidia branded it has 7.5 CUDA and not the latest drivers. I could not find an AMI with CUDA 8.0 without a bunch of other stuff and a massive boot drive.
If you “connect” the drive to an existing server then you can use any AMI you like as nothing is installed with this method. However all of the apps.py is written for amazon ami which is centos based so uses yum not apt-get. Also the launch of an instance installs docker and nvidia-docker using yum. So that will not work on ubuntu say. But then why do you care what the AMI is - if you use docker you can install any image you like in that? You can also install any programs you want in the fastai container. You just “docker exec -it fastai bash”. This opens a bash console in fastai container. You can then install anything you want.
Finally…the home drectory in the docker container is mapped to /v1 which is the root of the external volume. The jupyter config is stored there. So you can change the password in plain text. Also the notebooks are copied there. So you have a master copy of the notebooks inside the docker container; and a copy on /v1, the root of the external volume.