Code and data to support the paper -- Remember to forget: A study on verbatim memorization of literature in Large Language Models
This study probes the memorisation of books in English and French of open-source language models, specifically examining the extent to which these models have internalized English and French literature. Utilizing the name cloze inference methodology pioneered by Chang et al., we extend the analysis to evaluate the models' capacity for verbatim recall of literary texts in French.