CPPSourceGen Python Package

Summary

The CPPSource package is used to generate a C++ source file for executing the distributed inference. The source file generation depends on the mapping found in the model summary. The C++ source file generation relies on the Jinja package to read a template and populate its missing sections with the appropriate pieces of C++ code.

generate Script

Path: generate.py

The script can be used to generate the C++ source file by reading the directory where the different single-layer models are saved and generate source files for running inference of every layer on C++ on the CPU, GPU and TPU. Note: The directory can only be created and populated after running the split.py Script. The generated Source files should be saved under the following directory.

usage: 
python3 generate.py -md <path/to/tflite/single/layer/models/directory> 


Arguments:
  -md     --ModelsDirectory                 Directory where the single-layer tflite models are saved
  -fm     --FinalMapping                    Path to the Model Summary