Retrieval from Within: An Intrinsic Capability of Attention-Based Models
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_ | 1866911660728385536 |
|---|---|
| author | Hoffer, Elad Blau, Yochai Kinderman, Edan Banner, Ron Soudry, Daniel Ginsburg, Boris |
| author_facet | Hoffer, Elad Blau, Yochai Kinderman, Edan Banner, Ron Soudry, Daniel Ginsburg, Boris |
| contents | Retrieval-augmented generation (RAG) typically treats retrieval and generation as separate systems. We ask whether an attention-based encoder-decoder can instead retrieve directly from its own internal representations. We introduce INTRA (INTrinsic Retrieval via Attention), a framework where decoder attention queries score pre-encoded evidence chunks that are then directly reused as context for generation. By construction, INTRA unifies retrieval and generation, eliminating the retriever-generator mismatch typical of RAG pipelines. This design also amortizes context encoding by reusing precomputed encoder states across queries. On question-answering benchmarks, INTRA outperforms strong engineered retrieval pipelines on both evidence recall and end-to-end answer quality. Our results demonstrate that attention-based models already possess a retrieval mechanism that can be elicited, rather than added as an external module. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_05806 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Retrieval from Within: An Intrinsic Capability of Attention-Based Models Hoffer, Elad Blau, Yochai Kinderman, Edan Banner, Ron Soudry, Daniel Ginsburg, Boris Machine Learning Retrieval-augmented generation (RAG) typically treats retrieval and generation as separate systems. We ask whether an attention-based encoder-decoder can instead retrieve directly from its own internal representations. We introduce INTRA (INTrinsic Retrieval via Attention), a framework where decoder attention queries score pre-encoded evidence chunks that are then directly reused as context for generation. By construction, INTRA unifies retrieval and generation, eliminating the retriever-generator mismatch typical of RAG pipelines. This design also amortizes context encoding by reusing precomputed encoder states across queries. On question-answering benchmarks, INTRA outperforms strong engineered retrieval pipelines on both evidence recall and end-to-end answer quality. Our results demonstrate that attention-based models already possess a retrieval mechanism that can be elicited, rather than added as an external module. |
| title | Retrieval from Within: An Intrinsic Capability of Attention-Based Models |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2605.05806 |