6/ Mitigating Memorization in LLMs – presents a modification of the next-token prediction objective called goldfish loss to help mitigate the verbatim generation of memorized training data.
Goldfish Loss: Mitigating Memorization in Large Language Models
By
–
