Stealing the decoding algorithms of language models

Date: March 15, 2024