Z-Ant

Project Overview

Zant (Zig-Ant) is an open-source SDK for deploying optimized neural networks (NNs) on microcontrollers.

Why Zant?

• Lack of DL Support: Devices like TI Sitara, Raspberry Pi Pico, or ARM Cortex-M lack comprehensive DL libraries.
• Open-source: End-to-end NN deployment and optimization open-source solution.
• Research-Inspired: Implements optimization inspired by MIT's Han Lab research.
• Academic Collaboration: Developed in collaboration with institutions like Politecnico di Milano.

Key Features

• Real-time Optimizations: Quantization, pruning, buffer optimization.
• Cross-platform Compatibility: ARM Cortex-M, RISC-V, and others.
• Modular and Easy Integration: Clear APIs, examples, and extensive documentation.

Use Cases

• Edge AI: Real-time anomaly detection, predictive maintenance.
• IoT & Autonomous Systems: Lightweight AI models for drones, robots, vehicles, IoT devices.

---

Roadmap

Short-Term Goals (Q1 2025)

• March 5: MNIST inference on Raspberry Pi Pico 2.
• April 30: Efficient YOLO on Raspberry Pi Pico 2.

Mid-Term Goals (Q2-Q3 2025)

• Shape Tracker implementation.
• Frontend GUI for library interaction.
• im2tensor for image preprocessing.
• Enhanced code generation (flash vs RAM execution).
• Expanded ONNX compatibility.

Long-Term Goals (Q3 2025)

• Advanced pruning and quantization support.
• Expanded microcontroller compatibility.
• Model execution benchmarking tools.
• Improved real-time inference capabilities.

Getting Started

Requirements

• Install the latest Zig compiler.
• Improve Zig proficiency via Ziglings.

Running the Project

zig build run

Running Tests

Add tests to build.zig/test_list.

• Regular tests:

  zig build test --summary all

• Heavy computational tests:

zig build test -Dheavy --summary all

Documentation

Follow Zig's doc-comments.

Using Zant

Generating Code for Models

zig build codegen -Dmodel=model_name [-Dlog -Duser_tests=user_tests.json]

Generated code will be placed in:

generated/model_name/
├── lib_{model_name}.zig
├── test_{model_name}.zig
└── user_tests.json

Testing Generated Models

zig build test-codegen -Dmodel=model_name

Integrating into your Project

Build the static library:

zig build lib -Dmodel=model_name -Dtarget={arch} -Dcpu={cpu}

Linking with CMake:

target_link_libraries(your_project PUBLIC path/to/libzant.a)

Logging (Optional)

To set a custom log function from your C code:

extern void setLogFunction(void (*log_function)(uint8_t *string));

Zig Build File (build.zig)

Key Build Commands:

• Standard build & test:

  zig build
  zig build test --summary all

• Run code generation:

  zig build codegen -Dmodel=model_name [-Dlog -Duser_tests=path/to/tests.json]

• Compile static library:

  zig build lib -Dmodel=model_name -Dtarget=target_arch -Dcpu=specific_cpu

• Generate onnx oneOperation models:

  zig build test-codegen-gen

• Build and Test generated onnx oneOperation models:

  zig build test-codegen

Build Options

• -Dtrace_allocator=true|false: Use tracing allocator for debugging.
• -Dlog=true|false: Enable detailed logging during code generation.
• -Duser_tests=path/to/user_tests.json: Specify custom tests.

CI/CD Pipeline

• We are committed to enhancing our Continuous Integration/Continuous Deployment (CI/CD) pipeline to ensure robustness, reliability, and performance of Zant across all supported platforms. Key improvements include: Hardware-in-the-Loop (HIL) Testing: Integrate a hardware test bench with connected microcontrollers (e.g., Raspberry Pi Pico, ARM Cortex-M) into the CI/CD pipeline to validate real-world performance and compatibility.

• Profiling in CI/CD: Automatically profile generated models during the pipeline to measure execution time, memory usage, and power consumption on target hardware.

• Daily Fuzzing Tests: Run fuzzing tests daily within the CI/CD pipeline to identify edge-case bugs and ensure model stability under unexpected inputs.

• Multi-Platform Build Matrix: Test builds across a variety of architectures (ARM, RISC-V) and configurations in parallel to catch platform-specific issues early.

• Automated Benchmarking: Include performance benchmarking in the pipeline to track inference speed and resource usage over time, ensuring optimizations don’t regress.

• Code Coverage Reporting: Generate and publish code coverage metrics with every CI run to maintain high test quality.

• Containerized CI Environment: Use Docker containers to standardize the CI/CD environment, ensuring consistent builds and tests across all contributors.

Containerization

• Follow our Docker guide.

Contributing

Join us on GitHub and shape the future of tinyML!

Contributors

All contributors. Let's grow together!