OCM

Operator Capability Manager for Tensorflow models support analysis on Intel Hardwares

Build

Requirements:

• Preferred TF version: 2.2
• Tensorflow C library
• Tensorflow for python

Step 1 - To Build Tensorflow (One Time Process)

• If you already have tensorflow c library and tensorflow python 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 TF C and Python packages and install python package in a virtual environment build_and_setup_tf.sh
#Note: building tensorflow could take around 1-1.5 hours

Step 2 - To Build OCM

• If the TF is built using step 1, then run,
  build_ocm.sh

• If tensorflow is installed in virtual environment, then activate it first and then run,
  build_ocm.sh ${TF_SRC_DIR}
# make sure the C library is present at following path ${TF_SRC_DIR}/bazel-bin/tensorflow/libtensorflow_cc.so.2

Steps to run the OCM C++ test application

• If Tensorflow for python is installed in virtual environment, then activate it
  cd build
./ov_ocm ${PATH_TO_TF_FROZEN_PB_FILE}