Transformer Memory as a Differentiable Search Index
Open accessAbstract
In this paper, we demonstrate that information retrieval can be accomplished with a single Transformer, in which all information about the corpus is encoded in the parameters of the model. To this end, we introduce the Differentiable Search Index (DSI), a new paradigm that learns a text-to-text model that maps string queries directly to relevant docids; in other words, a DSI model answers queries directly using only its parameters, dramatically simplifying the whole retrieval process. We study variations in how documents and their identifiers are represented, variations in training procedures, and the interplay between models and corpus sizes. Experiments demonstrate that given appropriate design choices, DSI significantly outperforms strong baselines such as dual encoder models. Moreover, DSI demonstrates strong generalization capabilities, outperforming a BM25 baseline in a zero-shot setup.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
DSI and its variations train pretrained language models on the target corpus and treat the model's parameters as an index of the corpus.
“For instance, DSI and its variations [134-137] have tried to train pretrained language models on the target corpus directly and then treat the model's parameter as a 'index' of the corpus.”
From Foundations of GenIR (2025) · cites this paper as [135]
Cites
Scaling Laws for Neural Language Models
Kaplan, J., McCandlish, S., Henighan, T., Brown, T. B., Chess, B., Child, R., Gray, S., Radford, A., Wu, J., & Amodei, D. (2020) · arXiv · Cited by 1,749 · cited 1× · see citing works
“Understanding the scaling behaviour of Transformers have garnered significant interest in recent years (Kaplan et al., 2020; Tay et al., 2021; Abnar et al., 2021).”
Background
Flagged first, then by citations
Cited by
Sorted by citations
Precise Zero-Shot Dense Retrieval without Relevance Labels
Gao, L., Ma, X., Lin, J., & Callan, J. (2023) · Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2023.acl-long.99 · Cited by 362 · see citing works
Is ChatGPT Good at Search? Investigating Large Language Models as Re-Ranking Agents
Sun, W., Yan, L., Ma, X., Wang, S., Ren, P., Chen, Z., Yin, D., & Ren, Z. (2023) · Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2023.emnlp-main.923 · Cited by 270 · see citing works
Dense Text Retrieval Based on Pretrained Language Models: A Survey
Zhao, W. X., Liu, J., Ren, R., & Wen, J.-R. (2024) · ACM Transactions on Information Systems · doi:10.1145/3637870 · Cited by 211 · see citing works
Continual Learning for Generative Retrieval over Dynamic Corpora
Chen, J., Zhang, R., Guo, J., de Rijke, M., Chen, W., Fan, Y., & Cheng, X. (2023) · Proceedings of the 32nd ACM International Conference on Information and Knowledge Management · doi:10.1145/3583780.3614821 · Cited by 40 · see citing works
Multiview Identifiers Enhanced Generative Retrieval
Li, Y., Yang, N., Wang, L., Wei, F., & Li, W. (2023) · Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2023.acl-long.366 · Cited by 37 · see citing works
Generative Retrieval for Book Search
Tang, Y., Zhang, R., Guo, J., de Rijke, M., Liu, S., Wang, S., Yin, D., & Cheng, X. (2025) · Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1 · doi:10.1145/3690624.3709435 · Cited by 5 · see citing works
Lightweight and Direct Document Relevance Optimization for Generative Information Retrieval
Mekonnen, K. A., Tang, Y., & de Rijke, M. (2025) · Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval · doi:10.1145/3726302.3730023 · Cited by 3 · see citing works
Memory in the Age of AI Agents
Hu, Y., Liu, S., Yue, Y., Zhang, G., Liu, B., Zhu, F., Lin, J., Guo, H., Dou, S., Xi, Z., et al. (2025) · arXiv · doi:10.48550/arxiv.2512.13564 · Cited by 1 · see citing works
Ai, Q., Zhan, J., & Liu, Y. (2025) · arXiv · doi:10.48550/arxiv.2501.02842
“…on this direction argue that, by training the neural models to encode the whole corpus, documents and information would be implicitly stored in the parameters of the models, and this parameter-based indexes have better storage efficiency than traditional term-based or vector-based indexes [135].”
- Authors
- Tay, Y., Tran, V., Dehghani, M., Ni, J., Bahri, D., Mehta, H., Qin, Z., Hui, K., Zhao, Z., Gupta, J., Schuster, T., Cohen, W., & Metzler, D.
- Venue
- Advances in Neural Information Processing Systems 35
- Published
- 2022
- Pages
- 21831-21843
- ISSN
- 1049-5258 (ISSN-L)
- OpenAlex
- W7133208784
- Cited by
- 128
- References
- 1
- Citations by year
- 2026: 20 · 2025: 28 · 2024: 34 · 2023: 38 · 2022: 8
- FWCI
- 14.93
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- neural information processing systems (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Information Systems › Information Retrieval and Search Behavior
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Image and Video Retrieval Techniques
- Keywords
- search index, memory-augmented transformer
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
