Understanding Aidge’s architecture

Modular by design, flexible by nature

Aidge is built on a robust and flexible modular architecture, a foundational pillar that sets it apart. This modularity isn’t just a technical feature; it’s crucial for Aidge’s customization, extensibility, and adaptability to the diverse and evolving demands of AI deployment.

Core principles driving our design:

• Plugin-driven: we engineered a lightweight core module designed for seamless expansion with easily integrated plug-ins.

• Dependency minimization: this approach simplifies integration and management, avoiding conflicts within your current workflows.

By breaking down the framework into distinct, interchangeable components, Aidge offers key advantages:

• Pick and choose: integrate only the functionalities you need.

• Customize freely: replace existing functionalities with your own implementations.

• Expand easily: add new capabilities without touching the core system.

• Code engagement: quickly understand, explore, and contribute to relevant parts of the codebase.

• Discover how easy it is to add an operator to the C++ export by doing our tutorial.

Module ecosystem overview

Category	Modules	Description
Core foundation	aidge_core	The essential building blocks for manipulating AI models and data, indispensable for all other modules.
Execution backends	aidge_backend_cpu aidge_backend_cuda	Enable Aidge to run on specific hardware devices like CPUs and NVIDIA GPUs, providing concrete implementations.
Export modules	aidge_export_cpp aidge_export_arm_cortexm aidge_export_tensorrt aidge_export_acetone aidge_export_openvx	Generate deployable code for various target environments, seamlessly integrating AI models into diverse systems.
Optimization suite	aidge_quantization aidge_compression aidge_pruning	Offer advanced techniques to optimize AI models, reducing size, improving speed, and minimizing resource consumption.
Interoperability layer	aidge_onnx aidge_interop_torch	Facilitate smooth interaction between Aidge and other popular AI frameworks (ONNX, PyTorch) for broad compatibility.
Development tools	aidge_model_explorer	Visualize and explore AI model architectures and data flow within Aidge for easier understanding and debugging.