arxiv.org
LLMZip: Lossless Text Compression using Large Language Models
We provide new estimates of an asymptotic upper bound on the entropy of English using the large language model LLaMA-7B as a predictor for the next token given a window of past tokens. This estimate i...