GGML models that can run f16 41.68 ms per token and q8 23.76 ms per token giving good results

Usage example

Trained model

Update README.md

initial commit