Zant (Zig-Ant) is an open-source SDK designed to simplify deploying Neural Networks (NN) on microprocessors. Written in Zig, Zant prioritizes cross-compatibility and efficiency, providing tools to import, optimize, and deploy NNs seamlessly, tailored to specific hardware.
- Many microcontrollers (e.g., ATMEGA, TI Sitara) lack robust deep learning libraries.
- No open-source solution exists for end-to-end NN optimization and deployment.
- Inspired by cutting-edge research (e.g., MIT Han Lab), we leverage state-of-the-art optimization techniques.
- Collaborating with institutions like Politecnico di Milano to advance NN deployment on constrained devices.
- Built for flexibility to adapt to new hardware without codebase changes.
- Optimized Performance: Supports quantization, pruning, and hardware acceleration (SIMD, GPU offloading).
- Efficient Memory Usage: Incorporates memory pooling, static allocation, and buffer optimization.
- Cross-Platform Support: Works on ARM Cortex-M, RISC-V, and more.
- Ease of Integration: Modular design with clear APIs, examples, and documentation.
- Real-Time Applications: Object detection, anomaly detection, and predictive maintenance on edge devices.
- IoT and Autonomous Systems: Enable AI in IoT, drones, robots, and vehicles with constrained resources.
- Install the latest Zig compiler.
- Brush up your Zig skills with Ziglings exercises.
Navigate to the project folder and execute:
zig build run- Add new test files to
build.zig/test_listif not already listed. - Run:
(Ignore stderr warnings.)
zig build zig build test_all --summary all
Generated using Zig's standard documentation format.
Follow the Docker Guide for containerized usage.
Contribute to Zant on GitHub. Let’s make NN deployment on microcontrollers efficient, accessible, and open!