goldfish loss
Be like a Goldfish, Don't Memorize! Mitigating Memorization in Generative LLMs
Hans, Abhimanyu, Wen, Yuxin, Jain, Neel, Kirchenbauer, John, Kazemi, Hamid, Singhania, Prajwal, Singh, Siddharth, Somepalli, Gowthami, Geiping, Jonas, Bhatele, Abhinav, Goldstein, Tom
To mitigate memorization, we introduce a subtle modification to the next-token training objective that we call the goldfish loss. During training, a randomly sampled subset of tokens are excluded from the loss computation. These dropped tokens are not memorized by the model, which prevents verbatim reproduction of a complete chain of tokens from the training set. We run extensive experiments training billion-scale Llama-2 models, both pre-trained and trained from scratch, and demonstrate significant reductions in extractable memorization with little to no impact on downstream benchmarks.