batch-shipyard/recipes/Caffe-CPU

Caffe-CPU

This recipe shows how to run Caffe on a single node using CPU only.

Configuration

Please see refer to this set of sample configuration files for this recipe.

Pool Configuration

The pool configuration should enable the following properties:

• max_tasks_per_node must be set to 1 or omitted

Other pool properties such as publisher, offer, sku, vm_size and vm_count should be set to your desired values.

Global Configuration

The global configuration should set the following properties:

• docker_images array must have a reference to a valid Caffe CPU-enabled Docker image. Although you can use the official BVLC/caffe Docker images, for this recipe the alfpark/caffe:cpu contains all of the required files and scripts to run the MNIST convolutional example.

Jobs Configuration

The jobs configuration should set the following properties within the tasks array which should have a task definition containing:

• docker_image should be the name of the Docker image for this container invocation, e.g., alfpark/caffe:cpu
• command should contain the command to pass to the Docker run invocation. For the alfpark/caffe:cpu Docker image and to run the MNIST convolutional example, we are using a run_mnist.sh helper script. Thus, the command would simply be: "/caffe/run_mnist.sh"

Dockerfile and supplementary files

The Dockerfile for the Docker image can be found here.

You must agree to the Caffe License prior to use.