pyzenith 0.1.4

Cross-Platform ML Optimization Framework with ONNX Interpreter

Zenith

Cross-Platform ML Optimization Framework

A model-agnostic and hardware-agnostic unification and optimization framework for Machine Learning.

Features

• Unified API for PyTorch, TensorFlow, JAX, and ONNX models
• Automatic graph optimizations (fusion, constant folding, dead code elimination)
• Multi-backend support (CPU, CUDA, ROCm, TPU)
• Mixed precision training and inference (FP16, BF16, INT8)
• Property-based testing with mathematical guarantees

Installation

# Basic installation
pip install pyzenith

# With framework support
pip install pyzenith[onnx,pytorch,tensorflow,jax]

# Development installation
pip install -e ".[dev]"

Quick Start

import zenith
from zenith.core import GraphIR, DataType, Shape, TensorDescriptor

# Create a computation graph
graph = GraphIR(name="my_model")
graph.add_input(TensorDescriptor("x", Shape([1, 3, 224, 224]), DataType.Float32))

# Apply optimizations
from zenith.optimization import PassManager
pm = PassManager()
pm.add("constant_folding")
pm.add("dead_code_elimination")
optimized = pm.run(graph)

Architecture

+-------------------------------------------------------------+
|                    Python User Interface                    |
+-------------------------------------------------------------+
|              Framework-Specific Adapters Layer              |
|          (PyTorch, TensorFlow, JAX -> ONNX -> IR)           |
+-------------------------------------------------------------+
|       Core Optimization & Compilation Engine (C++)          |
|  - High-Level Graph Optimizer & IR                          |
|  - Kernel Scheduler & Auto-Tuner                            |
|  - Mathematical Kernel Library                              |
+-------------------------------------------------------------+
|           Hardware Abstraction Layer (HAL)                  |
|              CPU (SIMD) | CUDA | ROCm | TPU                 |
+-------------------------------------------------------------+

Documentation

• API Reference
• Architecture
• Blueprint

Development

# Run tests
pytest tests/python/ -v

# Run with coverage
pytest tests/python/ --cov=zenith --cov-report=term-missing

# Security scan
bandit -r zenith/ -ll

Current Status

• Phase 4: Quality Assurance & Documentation
• 198 tests passing
• 66% code coverage
• 0 HIGH severity security issues

Author

Wahyu Ardiansyah - Lead Architect

License

Apache License 2.0 - See LICENSE