container.md

Getting Started with the ML Suite Docker Container

The ML Suite Docker container provides users with a largely self contained software environment for running inference on Xilinx FPGAs. The only external dependencies are:

• docker-ce
• Xilinx XRT 2018.2 (Xilinx Run Time)

1. Install Docker

   • ubuntu
   • centos

   Note: Ensure /var/lib/docker has sufficient space (Should be > 5GB), if not move your Docker Root elsewhere.

2. Download the appropriate ML Suite Container from xilinx.com

   • ML Suite TensorFlow Container
   • ML Suite Caffe Container

3. Load the appropriate container

   # May need sudo
   $ docker load < xilinx-ml-suite-ubuntu-16.04-xrt-2018.2-tensorflow-1.12.0-mls-1.5.tar.gz
   $ docker load < xilinx-ml-suite-ubuntu-16.04-xrt-2018.2-caffe-1.0-mls-1.5.tar.gz
   

4. Use the provided script to launch and interact with the container

   # May need sudo, use appropriate argument, which is the image tag, ex:
   $ ./docker_run.sh xilinx-ml-suite-ubuntu-16.04-xrt-2018.2-tensorflow-1.12.0-mls-1.5
   $ ./docker_run.sh xilinx-ml-suite-ubuntu-16.04-xrt-2018.2-caffe-1.0-mls-1.5
   
   # The script by default will launch a container with the --rm flag, meaning after exiting, the container will be removed. 
   # Changes to source files will be lost.
   # A directory $MLSUITE_ROOT/share will be mounted in the container and can be used for easy file transfer between container and host.
   
   

5. Follow the Jupyter notebook or command line examples in the container

   Jupyter Notebook Tutorials

   • TensorFlow Image Classification
   • Caffe Image Classification
   • Caffe Object Detection w/ YOLOv2
   • Caffe Object Detection w/ YOLOv3
     Command Line Examples
   • Caffe ImageNet Benchmark Models
   • Caffe VOC SSD Example
   • Deployment Mode Examples

6. Follow Container Pipeline Example for example on how to stitch a container pipeline for preparing the model and running an inference server