Ternary Neural Networks in C#

The trend of quantization in neural networks may soon evolve with the introduction of models that utilize ternary logic. This article delves into various training formats, ranging from FP32 to FP4, and examines the potential of these new models. We will also create a version of the Microsoft Bitnet 1.58 model using C#, which will be compact in size, reaching as low as 17.1 KB in its most extreme form. This exploration aims to highlight the advantages and challenges of implementing ternary logic in neural networks, paving the way for more efficient and powerful AI systems.