LUME: LLM Unlearning with Multitask Evaluations
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866913709690978304 |
|---|---|
| author | Ramakrishna, Anil Wan, Yixin Jin, Xiaomeng Chang, Kai-Wei Bu, Zhiqi Vinzamuri, Bhanukiran Cevher, Volkan Hong, Mingyi Gupta, Rahul |
| author_facet | Ramakrishna, Anil Wan, Yixin Jin, Xiaomeng Chang, Kai-Wei Bu, Zhiqi Vinzamuri, Bhanukiran Cevher, Volkan Hong, Mingyi Gupta, Rahul |
| contents | Unlearning aims to remove copyrighted, sensitive, or private content from large language models (LLMs) without a full retraining. In this work, we develop a multi-task unlearning benchmark (LUME) which features three tasks: (1) unlearn synthetically generated creative short novels, (2) unlearn synthetic biographies with sensitive information, and (3) unlearn a collection of public biographies. We further release two fine-tuned LLMs of 1B and 7B parameter sizes as the target models. We conduct detailed evaluations of several recently proposed unlearning algorithms and present results on carefully crafted metrics to understand their behavior and limitations. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_15097 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | LUME: LLM Unlearning with Multitask Evaluations Ramakrishna, Anil Wan, Yixin Jin, Xiaomeng Chang, Kai-Wei Bu, Zhiqi Vinzamuri, Bhanukiran Cevher, Volkan Hong, Mingyi Gupta, Rahul Computation and Language Machine Learning Unlearning aims to remove copyrighted, sensitive, or private content from large language models (LLMs) without a full retraining. In this work, we develop a multi-task unlearning benchmark (LUME) which features three tasks: (1) unlearn synthetically generated creative short novels, (2) unlearn synthetic biographies with sensitive information, and (3) unlearn a collection of public biographies. We further release two fine-tuned LLMs of 1B and 7B parameter sizes as the target models. We conduct detailed evaluations of several recently proposed unlearning algorithms and present results on carefully crafted metrics to understand their behavior and limitations. |
| title | LUME: LLM Unlearning with Multitask Evaluations |
| topic | Computation and Language Machine Learning |
| url | https://arxiv.org/abs/2502.15097 |