Zant v1.0.0

[v1.0.0] - 2025-08-30

Added

• One-click deployment and test shell scripts (./create MyModel ) for reproducible pipelines.
• ONNX preprocessing stage feeding Zant IR (layout normalization, graph cleanup, graph optimization).
• Graph-split testing pipeline (./zant onnx_extract --path path/my_model.onnx \ zig build extractor-* -Dmodel="my_model") validating per-node intermediate I/O.
• Expanded math operations and full CNN support (conv/pool/activations).
• New documentation: end-to-end guides and troubleshooting.

Changed

• CLI redesigned for faster workflows and consistent flags; improved help and diagnostics.
• Default conversion pipeline now routes through ONNX preprocessing.

Fixed

• Clearer error surfaces and exit codes across CLI/tooling.

Notes

• Rebuild artifacts with v1.0.0 to benefit from preprocessing and per-node tests.

V 0.1

Changelog

• Project Overview and Documentation
• Updated Zant description as an SDK for deploying optimized neural networks on microcontrollers
• Revised key features, use cases, and reasons to use Zant
• Updated roadmap with short-term (Q1 2025), mid-term (Q2-Q3 2025), and long-term (Q3 2025) goals
• Added comprehensive documentation for code generation and model integration
• Added instructions for generating code for models and testing them
• Added section on integrating projects with Zant via static library and CMake
• Build System and Dependencies
• Updated Zig version from 0.13.0 to 0.14.0
• Removed build.zig.zon file
• Added new module for code generation in build.zig
• Added executable for code generation with options
• Added test for oneOp models with dependencies

Code Generation

• Renamed src/codeGen directory to src/CodeGen for consistency
• Added shape_handler and zant_codegen modules
• Added functions for parsing input shapes from codegen options
• Added network output initialization handling
• Added graph serialization and code generation functionality
• Added test file generation capabilities
• Added support for exporting predict functions

Tensor Math Operations

• Completely restructured tensor math operations into individual files
• Added implementations for activation functions:
• ReLU, Leaky ReLU, Sigmoid, Softmax, Tanh
• Added element-wise operations:
• Add, Subtract, Multiply, Divide, Ceil
• Added shape operations:
• Reshape, Resize, Transpose, Concat, Split, Slice, Unsqueeze, Identity, Neg
• Added convolution and matrix operations:
• Convolution, MatMul, Gemm
• Added reduction operations:
• ReduceMean
• Added padding and pooling operations

ONNX Support

• Added comprehensive ONNX model parsing capabilities
• Added structs for ONNX components:
• ModelProto, GraphProto, NodeProto, TensorProto, AttributeProto
• TensorShapeProto, ValueInfoProto, TypeProto, StringStringEntryProto
• TensorAnnotation, SparseTensorProto, Segment, DataLocation
• Added shape inference for ONNX operators

Testing

• Added user tests for MNIST and other models
• Added benchmarking for tensor operations
• Added Python scripts for generating test ONNX models
• Added tests for tensor math operations
• Added random data prediction tests
• Removed Components
• Removed data loading and processing functionality
• Removed neural network layer implementations (Dense, Convolutional, etc.)
• Removed model training functionality
• Removed model import/export functionality
• Removed optimizer implementations
• Removed loss function implementations
• Removed tests for removed components

Other Changes

• Updated GitHub issue templates with new contact links
• Added tensor math issue template with rules for writing methods
• Updated GitHub workflows to run on "main", "feature", and "codegen" branches
• Updated .gitignore rules for callgrind files and results.json