📐 Fast token estimation at 96% accuracy of a full tokenizer in a 2kB bundle - View it on GitHub