TMNN is a Python library for building custom neural networks from scratch, designed for consumer computing on CPU. With a focus on ease-of-use and efficiency, it aims to enable the creation of smaller, cheaper, and faster models without compromising on intelligence or capability.
- Compact and straightforward API for constructing neural networks.
- Optimized for CPU-based consumer-grade hardware.
- Future plans for OpenGL/CL and Vulkan support for general GPU compatibility.
The primary objective is to create smart, fast, and capable chat-like models that are quantized for optimal performance. TMNN is not meant to replace or compete with larger and more comprehensive frameworks but to provide a simpler, quicker avenue for neural network development and deployment on consumer hardware.