Be like a Goldfish, Don't Memorize! Mitigating Memorization in Generative LLMs
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866909375122112512 |
|---|---|
| author | Hans, Abhimanyu Wen, Yuxin Jain, Neel Kirchenbauer, John Kazemi, Hamid Singhania, Prajwal Singh, Siddharth Somepalli, Gowthami Geiping, Jonas Bhatele, Abhinav Goldstein, Tom |
| author_facet | Hans, Abhimanyu Wen, Yuxin Jain, Neel Kirchenbauer, John Kazemi, Hamid Singhania, Prajwal Singh, Siddharth Somepalli, Gowthami Geiping, Jonas Bhatele, Abhinav Goldstein, Tom |
| contents | Large language models can memorize and repeat their training data, causing privacy and copyright risks. To mitigate memorization, we introduce a subtle modification to the next-token training objective that we call the goldfish loss. During training, randomly sampled subsets 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_10209 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | 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 Computation and Language Large language models can memorize and repeat their training data, causing privacy and copyright risks. To mitigate memorization, we introduce a subtle modification to the next-token training objective that we call the goldfish loss. During training, randomly sampled subsets 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. |
| title | Be like a Goldfish, Don't Memorize! Mitigating Memorization in Generative LLMs |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2406.10209 |