Operator Capability Manager (OCM)

OCM checks Op support to run a Deep Learning model using OpenVINO on Intel Platforms (CPU, GPU, MYRIAD and HDDL). Currently it supports only Tensorflow models.

Build

Requirements:

• Tensorflow Version: 2.5.0
• Tensorflow CC library
• Tensorflow for python

Step 1 - To Build Tensorflow (One Time Process)

• If you already have Tensorflow CC library and Tensorflow Framework library installed then move to step 2 to build OCM
• Setup bazel, follow the steps mentioned on the following link https://www.tensorflow.org/install/source#install_bazel
• Run the following bash file and it will build Tensorflow CC and Python packages and install python package in a virtual environment
  build_and_setup_tf.sh ${TF_SRC_DIR}
# if the Tensorflow source code is already cloned at ${TF_SRC_DIR} then it will checkout the required version and will rebuild it
# otherwise it will clone Tensorflow source code at ${TF_SRC_DIR} path and will build it
# Note: building tensorflow could take several hours based on system config

Step 2 - To Build OCM

• Pre-requisite - Tensorflow Framework and CC libraries i.e. libtensorflow_framework.so and libtensorflow_cc.so
• If tensorflow is built and setup using Step 1 then activate the already created virtual environment, source ${TF_SRC_DIR}/ocm_venv/bin/activate
• Otherwise Make sure that the current python environment has Tensorflow of required version pre-installed build_ocm.sh ${TF_SRC_DIR}
# make sure Tensorflow CC library(libtensorflow_cc.so) is present at this path ${TF_SRC_DIR}/tensorflow/bazel-bin/tensorflow/

Steps to run the OCM C++ test application

cd build
./ov_ocm ${PATH_TO_TF_FROZEN_PB_FILE} ${DEVICE} {OPENVINO_VERSION}

Usage Examples:

./ov_ocm test.pb CPU 2021.2
./ov_ocm test.pb MYRIAD 2021.1