EmbeddedTorch

Convert your PyTorch models to embedded C++ code

EmbeddedTorch is a powerful Python library that allows you to convert PyTorch neural network models into optimized C++ code for deployment in embedded systems, microcontrollers, and resource-constrained environments.

Key Features

• 🚀 Easy Conversion: Convert PyTorch models to C++ with simple Python API
• 📦 Comprehensive Layers: Support for linear, convolutional, pooling, normalization, and activation layers
• ⚙️ Operations: Rich set of mathematical and tensor operations
• 🎯 Type-Safe: Support for different data types (float32, float64, etc.)
• 🔧 Flexible: Custom layer definitions and model composition support
• 🏗️ Production-Ready: Generate clean, optimized C++ code

Quick Start

from embeddedtorch.layers import EmbaeddableModel, LinearLayer
import torch

# Create a model
model = EmbaeddableModel(torch.float32)
model.add_layer(LinearLayer(10, 5, dtype=torch.float32))
model.add_layer(LinearLayer(5, 2, dtype=torch.float32))

# Generate C++ code
write_dep()
with open("out/code.cpp", "w", encoding="utf-8") as f:
    print(cpp_code(model.list), file=f)

Documentation

• Getting Started - Installation and basic usage
• API Reference - Detailed API documentation
• Layers - Available layer types and usage
• Operations - Mathematical operations
• Examples - Complete working examples

Get Started

Ready to convert your PyTorch models to C++? Check out the Getting Started guide.

Installation

pip install embeddedtorch

License

MIT License - See LICENSE file for details