synap-python 0.9.0

Python bindings for SyNAP

SyNAP-Python-API

NPU accelerated inference with Python

Hardware   |   Models   |   Documentation

---

Overview

The SyNAP Python API provides Python bindings that closely mirror our SyNAP C++ API, enabling seamless integration for inference in Python. The Python bindings offer a straightforward yet flexible approach to deep learning workflows, allowing users to incorporate custom pre-processing and post-processing functionalities.

[!NOTE] SyNAP Python API serves as the primary interface for low-level access to SyNAP C++ Framework, enabling seamless integration for inference in Python.

[!TIP] For a rapid AI development experience, we recommend using SynapRT, which provides ready-to-use AI pipelines.

Available Modules and Classes

The SyNAP Python API provides access to the following classes and functions:

Core Module (synap)

• Network
• Tensors
• Tensor

Preprocessing Module (synap.preprocess)

• Preprocessor
• InputData

Postprocessing Module (synap.postprocess)

• Detector
• Classifier
• (Additional auxiliary helper classes)

Data Type Definitions (synap.types)

• DataType
• Dim2d
• Landmark
• Layout
• Mask
• Rect
• Shape

Building the Python Wheel

Follow the steps below to set up your development environment and build the Python wheel. You can build this on a Linux machine, WSL2, or locally on the Astra board.

Prerequisites

Ensure you have necessary build dependencies installed:

Linux/WSL2

sudo apt update
sudo apt install python3.10 python3.10-venv python3.10-dev
sudo apt install -y build-essential cmake ninja-build

Astra

The python3-dev debian package for Astra is required. Install with:

wget https://synaptics-synap.github.io/examples-prebuilts/packages/python3-dev_3.10.13-r0_arm64.deb
dpkg -i python3-dev_3.10.13-r0_arm64.deb

Setup Virtual Environment

Create and activate a virtual environment to manage dependencies:

python3.10 -m venv venv
source venv/bin/activate

Build Package:

To build the Python wheel, just run:

./build.sh 

Use --verbose for detailed build steps and --clean for a clean build.

The build process generates a wheel file stored in the dist folder. The output should resemble the following:

Successfully built synap_python-0.9.0-cp310-cp310-manylinux_2_35_aarch64.whl
Building Python extensions ... Success!
Build completed successfully, wheel located at /home/../synap-python/dist/