NeuralDB: Scaling Knowledge Editing in LLMs to 100,000 Facts with Neural KV Database
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_ | 1866915407736078336 |
|---|---|
| author | Fei, Weizhi Shi, Hao Xu, Jing Peng, Jingchen Li, Jiazheng Zhang, Jingzhao Bai, Bo Han, Wei Chen, Zhenyuan Niu, Xueyan |
| author_facet | Fei, Weizhi Shi, Hao Xu, Jing Peng, Jingchen Li, Jiazheng Zhang, Jingzhao Bai, Bo Han, Wei Chen, Zhenyuan Niu, Xueyan |
| contents | Efficiently editing knowledge stored in large language models (LLMs) enables model updates without large-scale training. One possible solution is Locate-and-Edit (L\&E), allowing simultaneous modifications of a massive number of facts. However, such editing may compromise the general abilities of LLMs and even result in forgetting edited facts when scaling up to thousands of edits. In this paper, we model existing linear L\&E methods as querying a Key-Value (KV) database. From this perspective, we then propose NeuralDB, an editing framework that explicitly represents the edited facts as a neural KV database equipped with a non-linear gated retrieval module, % In particular, our gated module only operates when inference involves the edited facts, effectively preserving the general abilities of LLMs. Comprehensive experiments involving the editing of 10,000 facts were conducted on the ZsRE and CounterFacts datasets, using GPT2-XL, GPT-J (6B) and Llama-3 (8B). The results demonstrate that NeuralDB not only excels in editing efficacy, generalization, specificity, fluency, and consistency, but also preserves overall performance across six representative text understanding and generation tasks. Further experiments indicate that NeuralDB maintains its effectiveness even when scaled to 100,000 facts (\textbf{50x} more than in prior work). |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_18028 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | NeuralDB: Scaling Knowledge Editing in LLMs to 100,000 Facts with Neural KV Database Fei, Weizhi Shi, Hao Xu, Jing Peng, Jingchen Li, Jiazheng Zhang, Jingzhao Bai, Bo Han, Wei Chen, Zhenyuan Niu, Xueyan Computation and Language Artificial Intelligence Efficiently editing knowledge stored in large language models (LLMs) enables model updates without large-scale training. One possible solution is Locate-and-Edit (L\&E), allowing simultaneous modifications of a massive number of facts. However, such editing may compromise the general abilities of LLMs and even result in forgetting edited facts when scaling up to thousands of edits. In this paper, we model existing linear L\&E methods as querying a Key-Value (KV) database. From this perspective, we then propose NeuralDB, an editing framework that explicitly represents the edited facts as a neural KV database equipped with a non-linear gated retrieval module, % In particular, our gated module only operates when inference involves the edited facts, effectively preserving the general abilities of LLMs. Comprehensive experiments involving the editing of 10,000 facts were conducted on the ZsRE and CounterFacts datasets, using GPT2-XL, GPT-J (6B) and Llama-3 (8B). The results demonstrate that NeuralDB not only excels in editing efficacy, generalization, specificity, fluency, and consistency, but also preserves overall performance across six representative text understanding and generation tasks. Further experiments indicate that NeuralDB maintains its effectiveness even when scaled to 100,000 facts (\textbf{50x} more than in prior work). |
| title | NeuralDB: Scaling Knowledge Editing in LLMs to 100,000 Facts with Neural KV Database |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2507.18028 |