deephdc/deep-oc-benchmarks_cnn repository overview

⁠DEEP-OC-benchmarks_cnn

This is a container that will run the DEEP as a Service API component, DEEPaaS API⁠, with tf_cnn_benchmarks⁠ from the TensorFlow team. The source code for integration of tf_cnn_benchmarks⁠ with DEEPaaS API⁠ is located in benchmarks_cnn_api⁠.

The application is divided into two main 'types':

• 'benchmark' - for easy benchmarking of computing resources
• 'pro' - to perform customizable training of neural networks (neural network model, batch_size, weight_decay, etc).

The 'benchmark' type is further subdivided into following 'flavors':

• 'synthetic': generated in-memory data mimicking ImageNet dataset, avoids storage I/O
• 'dataset': ca. 5GB subset of real ImageNet data is used (downloaded automatically but may take time!), therefore involves I/O. Useful in comparison with the 'synthetic' flavor.

'benchmark' type has as inputs: 'flavor' and the number of GPUs only, the rest is defined inside the application: Both flavors run sequentially 100 batches of googlenet, inception3, resnet50, vgg16 and sum 'average_examples_per_sec' to derive the final 'score'. The optimizer is set to 'sgd'. The batch size is defined per GPU and scaled with the GPU memory. Initial batch_sizes are set for 4GB GPU memory as:

• googlenet: 96
• inception3: 24
• resnet50: 24
• vgg16: 16

It is also possible to run both flavors on CPU but the batch_size is fixed to 16 for all neural networks independent of the memory available.

⁠Running the container

⁠Directly from Docker Hub

To run the Docker container directly from Docker Hub and start using the API simply run the following command:

$ docker run -ti -p 5000:5000 -p 6006:6006 deephdc/deep-oc-benchmarks_cnn:type

This command will pull the Docker container from the Docker Hub deephdc⁠ repository and start the default command (deepaas-run --listen-ip=0.0.0.0).

⁠Running via docker-compose

docker-compose.yml allows you to run the application with various configurations via docker-compose.

N.B! docker-compose.yml is of version '2.3', one needs docker 17.06.0+ and docker-compose ver.1.16.0+, see https://docs.docker.com/compose/install/⁠

If you want to use Nvidia GPU, you need nvidia-docker and docker-compose ver1.19.0+ , see nvidia/FAQ⁠

⁠Building the container

If you want to build the container directly in your machine (because you want to modify the Dockerfile for instance) follow the following instructions:

Building the container:

1. Get the DEEP-OC-benchmarks_cnn repository:

   $ git clone https://git.scc.kit.edu/deep/DEEP-OC-benchmarks_cnn
   

   Copy

2. Build the container:

   $ cd DEEP-OC-benchmarks_cnn
   $ docker build -t deephdc/deep-oc-benchmarks_cnn .
   

   Copy

3. Run the container (if you enable JupyterLab during the build, --build-arg jlab=true, you should also add port 8888, i.e. -p 8888:8888)::

   $ docker run -ti -p 5000:5000 -p 6006:6006 deephdc/deep-oc-benchmarks_cnn
   

   Copy

These three steps will download the repository from GitHub and will build the Docker container locally on your machine. You can inspect and modify the Dockerfile in order to check what is going on. For instance, you can pass the --debug=True flag to the deepaas-run command, in order to enable the debug mode.

⁠Connect to the API

Once the container is up and running, browse to http://localhost:5000 to get the OpenAPI (Swagger)⁠ documentation page.