mvp_wrapper

MVP Python Wrapper

This allows for running the MVP-accelerated Tensorflow-Lite Micro (TFLM) kernels from a Python script.

This is useful as it allows for running .tflite model files in the MVP simulator and determining values such as:

• Required runtime memory (i.e. RAM)
• Number of MVP clock cycles required to execute model
• Which parts the model were not able to be accelerated

This wrapper is not used directly. Instead, it is loaded by the TfliteMicro Python package when specifying the accelerator = mvp.

Source Code

• Python wrapper - This makes the MVP-accelerated TFLM kernels accessible to Python
• MVP Tensorflow-Lite Micro Kernels - The MVP-accelerated Tensorflow-Lite Micro kernels. These run on the embedded device or in the MVP simulator
• Python API - Python package that loads the TFLM wrapper as well as this Python wrapper

Additional Links

• Python API documentation
• C++ Development documentation

Building the Wrapper

Pre-built

This wrapper comes pre-built when installing the YZLITE Python package, e.g.:

pip install yizhu-yzlite

Automatic Build

This wrapper is automatically built when installing from source, e.g.:

git clone https://github.com/ReRAM-Labs/yzlite.git
cd yzlite
pip install -e .

Manual build via YZLITE command

To manually build this wrapper, issue the YZLITE command:

yzlite build mvp_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 <yzlite repo root>/user_options.cmake and add:

yzlite_set(YZLITE_TARGET yzlite_mvp_wrapper)

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: yzlite_mvp_wrapper.