What is TensorPlay?

TensorPlay is a learner-friendly, DIY-ready deep learning framework designed for educational purposes and small-scale experiments.

Core Goals

1. Self-Contained: Minimal third-party dependencies.
2. Customizable: Easy to add new hardware support, operators, and differentiation modes.
3. Multi-Scene: Suitable for research, mobile deployment, and static graph acceleration.

Architecture

TensorPlay is divided into 4 core libraries:

• P10: The computation engine (Tensor + TensorImpl).
• TPX: The autograd engine (only differentiation).
• Stax: Static graph optimization and acceleration.
• NN: High-level neural network components.