Usage

GPUs

Nota bene you must explicitly request a quota increase.

Establish a baseline

Make your project known to gat.

gat create

Create a Dockerfile, e.g.,

FROM tensorflow/tensorflow
COPY ./train.py .
CMD python train.py

Test your Dockerfile with

gat run-local

Now run the task in Compute Engine.

gat run-remote

Tail the output of your task.

gat log -f

Inspect results in the newly created run-local and run-remote directories.

Creating experiments

Suppose we want to change the learning rate \(\eta\) to 0.3.

gat create eta0.3

Modify the code to effect the change, and rerun gat run-local or gat run-remote.

Navigating between experiments

To list all experiments,

gat list

To return to the baseline,

gat master

To return to an experiment,

gat eta0.3