Efficient LLM Moderation with Multi-Layer Latent Prototypes
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_ | 1866913178336624640 |
|---|---|
| author | Chrabąszcz, Maciej Szatkowski, Filip Wójcik, Bartosz Dubiński, Jan Trzciński, Tomasz Cygert, Sebastian |
| author_facet | Chrabąszcz, Maciej Szatkowski, Filip Wójcik, Bartosz Dubiński, Jan Trzciński, Tomasz Cygert, Sebastian |
| contents | Although modern LLMs are aligned with human values during post-training, robust moderation remains essential to prevent harmful outputs at deployment time. Existing approaches suffer from performance-efficiency trade-offs and are difficult to customize to user-specific requirements. Motivated by this gap, we introduce Multi-Layer Prototype Moderator (MLPM), a lightweight and highly customizable input moderation tool. We propose leveraging prototypes of intermediate representations across multiple layers to improve moderation quality while maintaining high efficiency. By design, our method adds negligible overhead to the generation pipeline and can be seamlessly applied to any model. MLPM achieves state-of-the-art performance on diverse moderation benchmarks and demonstrates strong scalability across model families of various sizes. Moreover, we show that it integrates smoothly into end-to-end moderation pipelines and further improves response safety when combined with output moderation techniques. Overall, our work provides a practical and adaptable solution for safe, robust, and efficient LLM deployment. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_16174 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Efficient LLM Moderation with Multi-Layer Latent Prototypes Chrabąszcz, Maciej Szatkowski, Filip Wójcik, Bartosz Dubiński, Jan Trzciński, Tomasz Cygert, Sebastian Machine Learning Artificial Intelligence Computation and Language Cryptography and Security Although modern LLMs are aligned with human values during post-training, robust moderation remains essential to prevent harmful outputs at deployment time. Existing approaches suffer from performance-efficiency trade-offs and are difficult to customize to user-specific requirements. Motivated by this gap, we introduce Multi-Layer Prototype Moderator (MLPM), a lightweight and highly customizable input moderation tool. We propose leveraging prototypes of intermediate representations across multiple layers to improve moderation quality while maintaining high efficiency. By design, our method adds negligible overhead to the generation pipeline and can be seamlessly applied to any model. MLPM achieves state-of-the-art performance on diverse moderation benchmarks and demonstrates strong scalability across model families of various sizes. Moreover, we show that it integrates smoothly into end-to-end moderation pipelines and further improves response safety when combined with output moderation techniques. Overall, our work provides a practical and adaptable solution for safe, robust, and efficient LLM deployment. |
| title | Efficient LLM Moderation with Multi-Layer Latent Prototypes |
| topic | Machine Learning Artificial Intelligence Computation and Language Cryptography and Security |
| url | https://arxiv.org/abs/2502.16174 |