Neural network inference in pure Odin

It achieves a better score 99 percent, then a fully connected neural network or SVM based on MNIST. Also, fast inference in us micro seconds.

A interesting concept, that trains the neural network at each epoch with 1.58 bits ternary states -1, 0, 1.

Pure Odin Qwen3 GGUF inference for Apple Silicon (CPU + Metal GPU)

Inference of SqueezeNet.onnx model on CUDA with Odin

Fast and efficient Odin inference for Qwen3 GGUF with the smallest (0.6B) at the fullest (FP32)