Tensorflow-Lite Micro Python Wrapper¶
This allows for accessing the Tensorflow-Lite Micro (TFLM) C++ interpreter from a Python script.
This is useful as it allows for executing .tflite model files from a Python script running on Windows/Linux (i.e. without requiring an embedded device).
This provides useful information about the .tflite
model such as:
Required working memory (i.e. RAM)
If any of the layers of the model are not supported by TFLM
This wrapper is made accessible to a Python script via the TfliteMicro Python API. This Python API loads the C++ Python wrapper shared library into the Python runtime.
Source Code¶
Python wrapper - This makes the Tensorflow-Lite Micro C++ library accessible to Python
Tensorflow-Lite Micro - This is the Tensorflow-Lite Micro C++ library plus some additional helpers to aid development (NOTE: The actual TFLM library is downloaded by the build scripts)
TfliteMicroModel - This is a helper C++ library to make interfacing to the TFLM library easier for applications
Python API - Python package that loads this C++ wrapper
Additional Links¶
Building the Wrapper¶
Pre-built¶
This wrapper comes pre-built when installing the MLTK Python package, e.g.:
pip install silabs-mltk
Automatic Build¶
This wrapper is automatically built when installing from source, e.g.:
git clone https://github.com/siliconlabs/mltk.git
cd mltk
pip install -e .
Manual build via MLTK command¶
To manually build this wrapper, issue the MLTK command:
mltk build tflite_micro_wrapper
Manual build via CMake¶
This wrapper can also be built via CMake using Visual Studio Code or the Command Line.
To build the wrapper, the build_options.cmake file needs to be modified.
Create the file <mltk repo root>/user_options.cmake
and add:
mltk_set(MLTK_TARGET mltk_tflite_micro_wrapper)
Note
You must remove this option and clean the build directory before building the example applications
Then configure the CMake project using the Window/Linux GCC toolchain and build the target: mltk_tflite_micro_wrapper
.