Model Unlearning Objectives Vary for Distinct Language Functions
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866914602489479168 |
|---|---|
| author | Atil, Berk Gupta, Vipul Passonneau, Rebecca J. |
| author_facet | Atil, Berk Gupta, Vipul Passonneau, Rebecca J. |
| contents | Large language models (LLMs) learn undesirable properties during pretraining, including dangerous knowledge and toxic text generation. Just as post-training uses different objectives to shape different behaviors, we argue that unlearning methods should be designed for the language function at issue. To study this, we consider two mechanistically distinct unlearning goals, dangerous-knowledge unlearning and toxicity unlearning. For dangerous knowledge, we introduce a cosine-based, meta-learned variant of RMU. For toxicity, we propose a multi-layer objective based on layer-specific probe directions. Across four open-source 7-8B models, our methods achieve strong results, based on distinct training objectives for the two types of unlearning. Overall, our results suggest that unlearning should be studied as a family of problems, analogous to the multiple types of LLM post-training. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_26454 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Model Unlearning Objectives Vary for Distinct Language Functions Atil, Berk Gupta, Vipul Passonneau, Rebecca J. Computation and Language Large language models (LLMs) learn undesirable properties during pretraining, including dangerous knowledge and toxic text generation. Just as post-training uses different objectives to shape different behaviors, we argue that unlearning methods should be designed for the language function at issue. To study this, we consider two mechanistically distinct unlearning goals, dangerous-knowledge unlearning and toxicity unlearning. For dangerous knowledge, we introduce a cosine-based, meta-learned variant of RMU. For toxicity, we propose a multi-layer objective based on layer-specific probe directions. Across four open-source 7-8B models, our methods achieve strong results, based on distinct training objectives for the two types of unlearning. Overall, our results suggest that unlearning should be studied as a family of problems, analogous to the multiple types of LLM post-training. |
| title | Model Unlearning Objectives Vary for Distinct Language Functions |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2605.26454 |