Fast inference from large lauguage models via speculative decoding - View it on GitHub