ZantFoundation/Z-Ant
Zant simplifies the deployment and optimization of neural networks on microprocessors
Prerequisites
# Clone and verify installation
git clone https://github.com/ZantFoundation/Z-Ant.git
cd Z-Ant
# - put your onnx model inside /datasets/models in a folder with the same of the model to to have: /datasets/models/my_model/my_model.onnx
# - simplify and prepare the model for zant inference engine
./zant input_setter --path /datasets/models/my_model/my_model.onnx --shape "your,model,sha,pe"
# - Generate test data
./zant user_tests_gen --model my_model
# --- GENERATING THE Single Node lib and test it ---
#For a N nodes model it creates N onnx models, one for each node with respective tests.
./zant onnx_extract --path /datasets/models/my_model/my_model.onnx
#generate libs for extracted nodes
zig build extractor-gen -Dmodel="my_model"
#test extracted nodes
zig build extractor-test -Dmodel="my_model"
# --- GENERATING THE LIBRARY and TESTS ---
# Generate code for a specific model
zig build lib-gen -Dmodel="my_model" -Denable_user_tests [-Ddynamic -Ddo_export -Dlog -Dcomm ... ]
# Test the generated code
zig build lib-test -Dmodel="my_model" -Denable_user_tests [-Ddynamic -Ddo_export -Dlog -Dcomm ... ]
# Build the static library
zig build lib -Dmodel="my_model" [-Dtarget=... -Dcpu=...]
IMPORTANT: see ZANT CLI for a better understanding and more details!
Command | What it does |
---|---|
zig build test |
Verify everything works |
zig build codegen -Dmodel=<name> |
Generate code from ONNX model |
zig build lib -Dmodel=<name> |
Build deployable static library |
zig build test-generated-lib -Dmodel=<name> |
Test your generated code |
Platform | Target Flag | CPU Examples |
---|---|---|
ARM Cortex-M | -Dtarget=thumb-freestanding |
-Dcpu=cortex_m33 , -Dcpu=cortex_m4 |
RISC-V | -Dtarget=riscv32-freestanding |
-Dcpu=generic_rv32 |
x86/Native | -Dtarget=native |
(auto-detected) |
Option | Description | Example |
---|---|---|
-Dmodel=<name> |
Your model name | -Dmodel=my_classifier |
-Dmodel_path=<path> |
Custom ONNX file | -Dmodel_path=models/custom.onnx |
-Dlog=true |
Enable detailed logging | -Dlog=true |
-Dcomm=true |
Add comments to generated code | -Dcomm=true |
Z-Ant includes Python scripts for ONNX model preparation:
# Prepare your model: set input shapes and infer all tensor shapes
./zant input_setter --path model.onnx --shape 1,3,224,224
# Generate test data for validation
./zant user_tests_gen --model model.onnx --iterations 10
# Create operator test models
./zant onnx_gen --op Conv --iterations 5
target_link_libraries(your_project PUBLIC path/to/libzant.a)
#include "lib_my_model.h"
// Optional: Set custom logging
extern void setLogFunction(void (*log_function)(uint8_t *string));
// Your inference code here
# Generate optimized library for image classifier
zig build codegen -Dmodel=mobilenet_v2 -Dmodel_path=models/mobilenet.onnx
zig build lib -Dmodel=mobilenet_v2 -Dtarget=thumb-freestanding -Dcpu=cortex_m33 -Doutput_path=deployment/
# Test on different architectures
zig build test-generated-lib -Dmodel=my_model -Dtarget=native
zig build test-generated-lib -Dmodel=my_model -Dtarget=thumb-freestanding -Dcpu=cortex_m4
# Run full test suite
zig build test --summary all
# Test heavy computational operations
zig build test -Dheavy=true
# Test specific operator implementations
zig build op-codegen-test -Dop=Conv
# Generate and test single operations
zig build op-codegen-gen -Dop=Add
Z-Ant/
โโโ src/ # Core source code
โ โโโ Core/ # Neural network core functionality
โ โโโ CodeGen/ # Code generation engine
โ โโโ ImageToTensor/ # Image preprocessing pipeline
โ โโโ onnx/ # ONNX model parsing
โ โโโ Utils/ # Utilities and helpers
โโโ tests/ # Comprehensive test suite
โโโ datasets/ # Sample models and test data
โโโ generated/ # Generated code output
โโโ examples/ # Arduino and microcontroller examples
โโโ docs/ # Documentation and guides
We welcome contributions from developers of all skill levels! Here's how to get involved:
All contributors are recognized in our Contributors list. Thank you for helping shape the future of tinyML!
This project is licensed under the LICENSE file in the repository.
Join us in revolutionizing AI on edge devices! ๐
GitHub โข Documentation โข Examples โข Community