text-generation-inference

5fa8ae04 - feat(server): optimize decode for sane tokenizers (#170)