Pith. sign in

REVIEW 2 major objections 4 minor 67 references

DistilVDR: A Compact End-to-End Visual Document Retriever via Dual-Student Distillation

T0 review · 2 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A 524M-parameter retriever distills an 8B visual-language teacher into a single vector per page, retaining 86.9% of its ViDoRe NDCG@5 while indexing 15.6 times smaller.

desk verdict A well-measured 524M VDR system with an unusually honest limitations section: useful as a deployable operating point, but the title's attribution to distillation is not yet proven because the from-scratch contrastive control was not run. read the letter →

arxiv 2608.10636 v1 pith:46OIZDUY submitted 2026-08-11 cs.IR cs.CLcs.CV

Zhuchenyang Liu , Ziyi Wang , Yao Zhang , Yu Xiao This is my paper · ORCID
classification cs.IRcs.CLcs.CV
keywords visualdocumentretrievaldual-studentdistillationsingle-vectorretrievercosinealignmentlossasymmetricencoderViDoRebenchmarkmodelcompressionembedding
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

DistilVDR sets out to show that a 524M-parameter, single-vector visual document retriever can inherit most of the retrieval quality of an 8B vision-language model by distilling both the query and document encoders onto that teacher's cached embedding space. The training objective is a pointwise cosine alignment loss, so no relevance labels, negative sampling, or contrastive term is needed at student-training time. On the 22-dataset ViDoRe suite the paper reports 61.74 average NDCG@5 for the high-resolution variant, 86.9% of the teacher's score and ahead of every reproduced sub-1B baseline; the compact variant scores 59.98 with a three-times-smaller visual-token budget. Both variants store one million documents in a 15.6-times-smaller index than the strongest sub-1B multi-vector baseline and index the corpus an order of magnitude faster. If true, the work turns a multi-billion-parameter visual retriever into a deployable 524M system that keeps most of the quality at a fraction of the index, memory, and latency cost.

What carries the argument

The load-bearing mechanism is dual-student distillation under a pointwise cosine alignment loss. For a document image d and a query q, the frozen 8B teacher produces cached targets T(d) and T(π◦q); each student minimizes Ld = 1 − ⟨fd(d), T(d)⟩ and Lq = 1 − ⟨fq(π◦q), T(π◦q)⟩ over L2-normalized vectors, with the two students trained independently and never sharing a forward pass. The document tower encodes dynamic tiles (up to six 448×448 tiles plus a thumbnail) with a 300M-parameter vision encoder, re-encodes the projected visual tokens with a 150M-parameter bidirectional text backbone, mean-pools, and projects to 4096 dimensions; the query tower is a 70M text encoder with the teacher's instruction prefix. Retrieval is a single dot product between the student vectors, and the teacher is discarded after training.

What would settle it

Train the identical 524M encoder-only architecture from scratch on the same 1.20M-image mixture with hard-negative mining and contrastive learning under a comparable compute budget; if that model matches or exceeds DistilVDR's 61.74 average NDCG@5, then the distillation objective itself is not the cause of the result. Alternatively, evaluate DistilVDR and the teacher on a corpus where the teacher's cosine ranking is known to be miscalibrated across modalities; if the student does not reproduce that miscalibration, the cosine-space assumption is doing less work than claimed.

Watch

Extended reading notes

Core claim

The paper's central claim is that a single-vector dense retriever at 524M parameters, trained by independently regressing each tower onto the frozen teacher's 4096-dimensional L2-normalized vectors, reproduces enough of the teacher's relevance geometry to dominate every reproduced sub-1B visual document retriever and to retain 86.9% of an 8B teacher's average NDCG@5. The authors argue the teacher's embedding space is a complete supervision signal because it was itself trained with relevance supervision, so the student objective can be purely geometric. They further claim the asymmetric encoder-only student, with 454M parameters on the document side and 70M on the query side, matches the text-query/image-document asymmetry of the task and that varying only the visual-tile budget yields two deployment points (HiRes and Fast) that trade 1.76 average NDCG@5 for a three-times-smaller visual-token budget. The gap decomposition shows the residual distance to the teacher is not additive across sides, and that the single-vector format itself is not the bottleneck since the teacher at 71.05 NDCG@5 is level with 4.4B multi-vector models.

Load-bearing premise

The student only learns to reproduce the teacher's per-item embedding directions, so if the teacher's cosine similarities do not order documents by relevance, or if matching each side separately throws away how query and document embeddings calibrate with each other, the student cannot recover teacher-level retrieval quality.

Editorial extensions

If this is right

  • Single-vector visual document retrieval at the 500M scale can retain roughly 85–87% of an 8B teacher's ViDoRe NDCG@5, closing much of the quality gap that previously forced multi-vector late interaction.
  • Deployments that need one dense vector per page get a 15.6-times-smaller index and roughly two-orders-of-magnitude-lower scoring latency than sub-1B multi-vector baselines, and indexing throughput rises by an order of magnitude.
  • Because the student objective needs no relevance labels, negatives, or contrastive term, the two towers can be trained independently and in parallel against cached teacher targets, and the teacher runs only once.
  • The visual-tile budget is a direct quality-versus-cost control: halving the tile budget from six to two tiles costs 1.76 average NDCG@5 while tripling throughput, which gives system builders a single knob for deployment.
  • The student is bounded by the teacher's own embedding quality: no side of the student can exceed the teacher on the side it replaces, so improving the teacher, or distilling from a multi-vector or cross-encoder teacher, is the path beyond the measured ceiling.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The flat contrastive-refinement result suggests that once a teacher's geometry is matched pointwise, explicit cross-modal coupling adds little; but the paper does not run the from-scratch contrastive control, so the claim that distillation beats contrastive training for this architecture remains untested.
  • Since teacher targets are cached once, sweeping teacher scale or family is cheap; a stronger single-vector teacher would likely lift both variants proportionally, and the same recipe could be dropped onto future embedding models without retraining the pipeline.
  • The fixed per-page tile budget ignores content complexity; a page-complexity estimate that allocates tiles to dense small-text pages could close part of the remaining v3 gap, which is concentrated exactly on such pages.
  • The 768-d Matryoshka prefix already costs only 3.19 average NDCG@5 while cutting the index 5.3x; combining it with scalar or product quantization, which the paper leaves untested, could make even the compact variant fit in a few GB per million pages.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper introduces DistilVDR, a 524M-parameter end-to-end visual document retriever obtained by distilling both the query and document encoders from a frozen 8B vision-language teacher (Qwen3-VL-Embedding-8B) under a pointwise cosine alignment loss. The student is asymmetric: a 70M text-only query encoder and a 454M document encoder that combines InternViT-300M, a dynamic tiling scheme, and a ModernBERT-base backbone. Two release variants differ only in the visual tile budget: HiRes (six tiles plus thumbnail) and Fast (two tiles plus thumbnail). The paper reports NDCG@5 on the 22-dataset ViDoRe v1+v2+v3 suite, with HiRes at 61.74 average (86.9% of the teacher) and Fast at 59.98, both exceeding every reproduced sub-1B baseline and both offering large savings in index footprint and indexing time relative to multi-vector baselines. The paper also includes a gap decomposition (teacher/student combinations), ablations over tile budget, data scale, output dimension, query backbone, and a contrastive-supervision refinement study, and it candidly lists several limitations, including the absence of a from-scratch contrastive control.

Significance. If the central claims hold, DistilVDR is a practically valuable system: it shows that a compact single-vector retriever can come close to an 8B teacher while reducing index storage by roughly 15.6× and indexing substantially faster than multi-vector baselines. The paper's strengths are its unified reproduction of twelve released retrievers under one evaluation and profiling protocol, its detailed gap decomposition, its careful reporting of training-data provenance and teacher precomputation cost, and its unusually honest limitations section. The release of code and the inclusion of a conservative attention-backend reference further support reproducibility. However, the causal attribution of the quality gain to distillation itself is not tested, and one efficiency claim is overstated; these issues require revision before the results can be fully accepted.

major comments (2)
  1. [§5.1 and Limitations] The paper's central claim, reflected in its title and framing, is that pointwise cosine distillation is what enables a 524M single-vector retriever to lead all reproduced sub-1B baselines. The Limitations section states: 'The control that would isolate distillation itself, the same 524M architecture trained contrastively from scratch on the same mixture with hard-negative mining and a comparable compute budget, is the principal experiment we did not run.' Section 5.1 starts from an already-distilled checkpoint and adds a contrastive term, so it cannot separate the contribution of the distillation objective from that of the architecture (InternViT-300M, tiling, ModernBERT), the 1.20M-image training mixture, and the compute budget. Given that the paper explicitly contrasts its route with from-scratch small models that 'struggle,' this missing control is load-bearing for the attribution. I recommend adding the from-scratch contrastive training of the same architecture on the same mixture, or, if that is infeasible, re-framing the claims to describe the full recipe rather than the distillation objective alone.
  2. [§4.3 and Abstract] The text claims that DistilVDR-Fast is 'an order of magnitude faster than every multi-vector baseline at any scale.' Table 3 reports SauerkrautLM-ColLFM2-450M, a multi-vector baseline, at 19.02 docs/sec, versus 99.04 docs/sec for DistilVDR-Fast; this is a speedup of approximately 5.2×, which is not an order of magnitude. The comparison is indeed >10× for the other sub-1B multi-vector baselines (colSmol-256M, colSmol-500M, ColModernVBERT), so the claim should be narrowed accordingly or reworded to 'several times faster than the fastest multi-vector baseline and an order of magnitude faster than the slower ones.' This overstatement appears in the abstract, Section 4.3, and the conclusion, so it should be corrected in all three places.
minor comments (4)
  1. [§4.2 / Table 1] The paper reports single-run NDCG@5 values without error bars or significance tests. Given that some reported differences are small (e.g., Fast versus HiRes on v1, 81.34 versus 82.81), providing multiple seeds or a variance estimate would strengthen the headline comparisons.
  2. [§4.2] The statement that 'HiRes leads every reproduced sub-1B baseline on the high-resolution-sensitive v3 benchmark' is accurate, but the 13.55-point margin over colSmol-500M should perhaps be contextualized by noting that the v3 benchmark is recent and the reproduced baselines may not have been trained or tuned on comparable v3-like professional-domain data.
  3. [Appendix G] The conservative attention-backend reference (eager attention) is a good transparency addition, but the main text would benefit from a sentence noting that even with eager attention, DistilVDR-Fast remains faster than every sub-1B multi-vector baseline, as the appendix reports.
  4. [§4.1] The teacher precomputation cost is reported in H200-GPU-hours, but the §4.1 sentence describing the three sharded batches is dense and would be clearer if the per-part hours were tied to the final deduplicated image counts in the same sentence.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: teacher targets are external and all headline numbers are held-out measurements.

full rationale

I find no circular step in the claimed derivation chain. The teacher (Qwen3-VL-Embedding-8B) is an externally released model trained with relevance supervision and is not tuned on ViDoRe in this paper; the students regress onto cached 4096-d teacher embeddings under the cosine alignment losses of Eqs. 1-2, and retrieval quality is then measured on held-out ViDoRe v1/v2/v3 benchmarks against externally released baselines. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, and no result is derived from an input that already contains the target quantity. The self-citations to NanoVDR supply the query training mixture and a loss-ablation comparison; neither is load-bearing for the headline claim. The query mixture is training data, not a benchmark result, and the NanoVDR ablation is cited only as a consistency check for the flat contrastive finding in Table 5, which is established by the paper's own experiments. The missing from-scratch contrastive control, candidly acknowledged in the Limitations section, is an attribution/completeness limitation rather than circularity: it means the margin over baselines cannot be uniquely attributed to distillation, but no reported number reduces by construction to a fitted constant or to a self-citation chain.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The method's main external inputs are the teacher's embedding space and pretrained backbone weights; no new physical or conceptual entities are introduced. The key free choice is Tmax, while the remaining recipe components are standard distillation hyperparameters.

free parameters (1)
  • Maximum visual tile count Tmax = 2 (Fast) or 6 (HiRes)
    Release-time design choice that distinguishes the two variants and directly controls the quality/efficiency trade-off. It is not fitted to validation data, but it is a hand-chosen free parameter of the system.
assumptions (4)
  • domain assumption The frozen teacher's L2-normalized 4096-d embedding space is a valid relevance metric for VDR.
    Section 3.5: the entire student objective regresses onto T(d) and T(pi q), so if teacher cosine similarities do not reflect relevance, the student inherits that failure.
  • domain assumption Pretrained InternViT-300M and ModernBERT-base weights transfer well to document-page visual tokens.
    Section 3.3: these backbones are selected for their pretraining and long-context properties, but the paper does not ablate alternative visual/text backbones at the same scale.
  • domain assumption Perceptual-hash deduplication at distance 0 removes all relevant train/eval overlap.
    Appendix B: dedup is performed offline with the imagehash library at perceptual-hash distance 0, which removes only exact visual duplicates; near-duplicate or rescaled pages could still leak.
  • domain assumption Independent cosine regressions on each side produce a jointly usable shared query/document space.
    Section 3.5 trains the two students with no cross-modal interaction; Section 5.1 tests adding contrastive coupling and finds it flat, but the assumption remains the basis of the deployed system.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DistilVDR: A Compact End-to-End Visual Document Retriever via Dual-Student Distillation." pith.science (2026). https://pith.science/paper/46OIZDUY

@misc{pith2026260810636,
  author       = {Pith},
  title        = {Pith review of: DistilVDR: A Compact End-to-End Visual Document Retriever via Dual-Student Distillation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/46OIZDUY}},
  note         = {Machine review of arXiv:2608.10636}
}
read the original abstract

Visual document retrieval (VDR) is dominated by multi-billion-parameter models that are slow to index at full corpus scale and expensive to serve. Prior compression routes either train a smaller multi-vector encoder from scratch or distil only the query side; neither yields a compact single-vector retriever end-to-end. We present DistilVDR, a 524M end-to-end VDR system distilled bilaterally from a single 8B vision-language teacher under a pointwise cosine alignment loss. All supervision comes from the frozen teacher's embedding space, which was itself trained with relevance supervision, so the student objective needs no relevance labels, negative sampling, or contrastive term. We match VDR's text-query and image-document input asymmetry with an asymmetric encoder-only student that concentrates visual capacity on the document side and keeps the query side at 70M parameters. We release two variants that share the same encoders and training and differ only in the document encoder's visual-tile budget: DistilVDR-HiRes attains 61.74 average NDCG@5 on ViDoRe v1+v2+v3 (86.9% of the 8B teacher) and leads every reproduced sub-1B baseline on the high-resolution-sensitive v3 benchmark, while DistilVDR-Fast attains 59.98 at a 3 times smaller visual-token budget. Both variants store one million documents in a 15.6 times smaller index than the strongest sub-1B multi-vector baseline and index the corpus an order of magnitude faster. The code is available at https://github.com/Ryenhails/NanoVDR.

Figures

Figures reproduced from arXiv: 2608.10636 by the authors.

Figure 1
Figure 1. Dual-student distillation architecture. The [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Per-epoch train (blue) and validation (red) loss [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

67 extracted references · 21 canonical work pages

  1. [1]

    International Conference on Learning Representations , volume=

    Faysse, Manuel and Sibille, Hugues and Wu, Tony and Omrani, Bilel and Viaud, Gautier and Hudelot, C. International Conference on Learning Representations , volume=

    work page

  2. [2]

    arXiv preprint arXiv:2505.17166 , year=

    Mac. arXiv preprint arXiv:2505.17166 , year=

    arXiv

  3. [3]

    arXiv preprint arXiv:2601.08620 , year=

    Loison, Ant. arXiv preprint arXiv:2601.08620 , year=

    arXiv

  4. [4]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages=

    Unifying multimodal retrieval via document screenshot embedding , author=. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages=

    work page 2024

  5. [5]

    arXiv preprint arXiv:2510.01149 , year=

    Teiletche, Paul and Mac. arXiv preprint arXiv:2510.01149 , year=

    arXiv

  6. [6]

    Liu, Zhuchenyang and Zhang, Yao and Xiao, Yu , journal=

    work page

  7. [7]

    Yu, Shi and Tang, Chaoyue and Xu, Bokai and Cui, Junbo and Ran, Junhao and Yan, Yukun and Liu, Zhenghao and Wang, Shuo and Han, Xu and Liu, Zhiyuan and others , booktitle=

    work page

  8. [8]

    2024 , howpublished =

    2024

Show all 67 references
  1. [9]

    2025 , howpublished =

    2025

  2. [10]

    2025 , howpublished =

    Tomoro-. 2025 , howpublished =

    2025

  3. [11]

    Li, Mingxin and Zhang, Yanzhao and Long, Dingkun and Chen, Keqin and Song, Sibo and Bai, Shuai and Yang, Zhibo and Xie, Pengjun and Yang, An and Liu, Dayiheng and others , journal=

  4. [12]

    2023 , eprint=

    Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond , author=. 2023 , eprint=

    2023

  5. [13]

    arXiv preprint arXiv:2409.12191 , year=

    Qwen2-vl: Enhancing vision-language model's perception of the world at any resolution , author=. arXiv preprint arXiv:2409.12191 , year=

    arXiv

  6. [14]

    arXiv preprint arXiv:2505.09388 , year=

    Qwen3 technical report , author=. arXiv preprint arXiv:2505.09388 , year=

    arXiv

  7. [15]

    Jiang, Ziyan and Meng, Rui and Yang, Xinyi and Yavuz, Semih and Zhou, Yingbo and Chen, Wenhu , journal=

  8. [16]

    2024 , organization=

    Wei, Cong and Chen, Yang and Chen, Haonan and Hu, Hexiang and Zhang, Ge and Fu, Jie and Ritter, Alan and Chen, Wenhu , booktitle=. 2024 , organization=

    2024

  9. [17]

    arXiv preprint arXiv:2010.11929 , year=

    An image is worth 16x16 words: Transformers for image recognition at scale , author=. arXiv preprint arXiv:2010.11929 , year=

    2010 arXiv

  10. [18]

    International conference on machine learning , pages=

    Learning transferable visual models from natural language supervision , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  11. [19]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Sigmoid loss for language image pre-training , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  12. [20]

    2502.14786 , archivePrefix=

    Michael Tschannen and Alexey Gritsenko and Xiao Wang and Muhammad Ferjad Naeem and Ibrahim Alabdulmohsin and Nikhil Parthasarathy and Talfan Evans and Lucas Beyer and Ye Xia and Basil Mustafa and Olivier Hénaff and Jeremiah Harmsen and Andreas Steiner and Xiaohua Zhai , year=....

    arXiv

  13. [21]

    Chen, Zhe and Wu, Jiannan and Wang, Wenhai and Su, Weijie and Chen, Guo and Xing, Sen and Zhong, Muyan and Zhang, Qinglong and Zhu, Xizhou and Lu, Lewei and others , booktitle=

  14. [22]

    arXiv preprint arXiv:2412.05271 , year=

    Expanding performance boundaries of open-source multimodal models with model, data, and test-time scaling , author=. arXiv preprint arXiv:2412.05271 , year=

    arXiv

  15. [23]

    Advances in neural information processing systems , volume=

    Visual instruction tuning , author=. Advances in neural information processing systems , volume=

  16. [24]

    International conference on machine learning , pages=

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models , author=. International conference on machine learning , pages=. 2023 , organization=

    2023

  17. [25]

    arXiv preprint arXiv:2504.05299 , year=

    Smolvlm: Redefining small and efficient multimodal models , author=. arXiv preprint arXiv:2504.05299 , year=

    arXiv

  18. [26]

    arXiv preprint arXiv:2408.12637 , year=

    Building and better understanding vision-language models: insights and future directions , author=. arXiv preprint arXiv:2408.12637 , year=

    arXiv

  19. [27]

    2025 , eprint=

    Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling , author=. 2025 , eprint=

    2025

  20. [28]

    Bert: Pre-training of deep bidirectional transformers for language understanding , author=. Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers) , pages=

    2019

  21. [29]

    Sanh, Victor and Debut, Lysandre and Chaumond, Julien and Wolf, Thomas , journal=

  22. [30]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Smarter, better, faster, longer: A modern bidirectional encoder for fast, memory efficient, and long context finetuning and inference , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  23. [31]

    arXiv preprint arXiv:1503.02531 , year=

    Distilling the knowledge in a neural network , author=. arXiv preprint arXiv:1503.02531 , year=

    arXiv

  24. [32]

    Advances in neural information processing systems , volume=

    Minilm: Deep self-attention distillation for task-agnostic compression of pre-trained transformers , author=. Advances in neural information processing systems , volume=

  25. [33]

    Proceedings of the 2020 conference on empirical methods in natural language processing (EMNLP) , pages=

    Dense passage retrieval for open-domain question answering , author=. Proceedings of the 2020 conference on empirical methods in natural language processing (EMNLP) , pages=

    2020

  26. [34]

    Sentence-bert: Sentence embeddings using siamese bert-networks , author=. Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP) , pages=

    2019

  27. [35]

    Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =

    Khattab, Omar and Zaharia, Matei , title =. Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2020 , isbn =. doi:10.1145/3397271.3401075 , abstract =

    2020

  28. [36]

    C ol BERT v2: Effective and Efficient Retrieval via Lightweight Late Interaction

    Santhanam, Keshav and Khattab, Omar and Saad-Falcon, Jon and Potts, Christopher and Zaharia, Matei. C ol BERT v2: Effective and Efficient Retrieval via Lightweight Late Interaction. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computa...

    2022 doi

  29. [37]

    Proceedings of the 31st ACM International Conference on Information & Knowledge Management , pages =

    Santhanam, Keshav and Khattab, Omar and Potts, Christopher and Zaharia, Matei , title =. Proceedings of the 31st ACM International Conference on Information & Knowledge Management , pages =. 2022 , isbn =. doi:10.1145/3511808.3557325 , abstract =

    2022

  30. [38]

    2020 , eprint=

    Approximate Nearest Neighbor Negative Contrastive Learning for Dense Text Retrieval , author=. 2020 , eprint=

    2020

  31. [39]

    R ocket QA : An Optimized Training Approach to Dense Passage Retrieval for Open-Domain Question Answering

    Qu, Yingqi and Ding, Yuchen and Liu, Jing and Liu, Kai and Ren, Ruiyang and Zhao, Wayne Xin and Dong, Daxiang and Wu, Hua and Wang, Haifeng. R ocket QA : An Optimized Training Approach to Dense Passage Retrieval for Open-Domain Question Answering. Proceedings of the 2021 Confe...

    2021 doi

  32. [40]

    Efficiently Teaching an Effective Dense Retriever with Balanced Topic Aware Sampling , year =

    Hofst\". Efficiently Teaching an Effective Dense Retriever with Balanced Topic Aware Sampling , year =. Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. doi:10.1145/3404835.3462891 , abstract =

  33. [41]

    R etro MAE : Pre-Training Retrieval-oriented Language Models Via Masked Auto-Encoder

    Xiao, Shitao and Liu, Zheng and Shao, Yingxia and Cao, Zhao. R etro MAE : Pre-Training Retrieval-oriented Language Models Via Masked Auto-Encoder. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022. doi:10.18653/v1/2022.emnlp-main.35

    2022 doi

  34. [42]

    2024 , eprint=

    Text Embeddings by Weakly-Supervised Contrastive Pre-training , author=. 2024 , eprint=

    2024

  35. [43]

    Multilingual

    Liang Wang and Nan Yang and Xiaolong Huang and Linjun Yang and Rangan Majumder and Furu Wei , year=. Multilingual. 2402.05672 , archivePrefix=

    arXiv

  36. [44]

    M 3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation

    Chen, Jianlyu and Xiao, Shitao and Zhang, Peitian and Luo, Kun and Lian, Defu and Liu, Zheng. M 3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation. Findings of the Association for Computational Linguistics:...

    2024 doi

  37. [45]

    jina-embeddings-v3: Multilingual Embeddings With Task

    Saba Sturua and Isabelle Mohr and Mohammad Kalim Akram and Michael Günther and Bo Wang and Markus Krimmel and Feng Wang and Georgios Mastrapas and Andreas Koukounas and Nan Wang and Han Xiao , year=. jina-embeddings-v3: Multilingual Embeddings With Task. 2409.10173 , archivePrefix=

    arXiv

  38. [46]

    Matryoshka Representation Learning , url =

    Kusupati, Aditya and Bhatt, Gantavya and Rege, Aniket and Wallingford, Matthew and Sinha, Aditya and Ramanujan, Vivek and Howard-Snyder, William and Chen, Kaifeng and Kakade, Sham and Jain, Prateek and Farhadi, Ali , booktitle =. Matryoshka Representation Learning , url =

  39. [47]

    Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , pages =

    Xu, Yiheng and Li, Minghao and Cui, Lei and Huang, Shaohan and Wei, Furu and Zhou, Ming , title =. Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , pages =. 2020 , isbn =. doi:10.1145/3394486.3403172 , abstract =

    2020

  40. [48]

    Proceedings of the 30th ACM International Conference on Multimedia , pages =

    Huang, Yupan and Lv, Tengchao and Cui, Lei and Lu, Yutong and Wei, Furu , title =. Proceedings of the 30th ACM International Conference on Multimedia , pages =. 2022 , isbn =. doi:10.1145/3503161.3548112 , abstract =

    2022

  41. [49]

    OCR -Free Document Understanding Transformer

    Kim, Geewook and Hong, Teakgyu and Yim, Moonbin and Nam, JeongYeon and Park, Jinyoung and Yim, Jinyeong and Hwang, Wonseok and Yun, Sangdoo and Han, Dongyoon and Park, Seunghyun. OCR -Free Document Understanding Transformer. Computer Vision -- ECCV 2022. 2022

    2022

  42. [50]

    2023 , editor =

    Lee, Kenton and Joshi, Mandar and Turc, Iulia Raluca and Hu, Hexiang and Liu, Fangyu and Eisenschlos, Julian Martin and Khandelwal, Urvashi and Shaw, Peter and Chang, Ming-Wei and Toutanova, Kristina , booktitle =. 2023 , editor =

    2023

  43. [51]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Tang, Zineng and Yang, Ziyi and Wang, Guoxin and Fang, Yuwei and Liu, Yang and Zhu, Chenguang and Zeng, Michael and Zhang, Cha and Bansal, Mohit , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2023 , pages =

    2023

  44. [52]

    2021 , eprint=

    BEIR: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models , author=. 2021 , eprint=

    2021

  45. [53]

    MTEB : Massive Text Embedding Benchmark

    Muennighoff, Niklas and Tazi, Nouamane and Magne, Loic and Reimers, Nils. MTEB : Massive Text Embedding Benchmark. Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics. 2023. doi:10.18653/v1/2023.eacl-main.148

    2023 doi

  46. [54]

    2018 , eprint=

    MS MARCO: A Human Generated MAchine Reading COmprehension Dataset , author=. 2018 , eprint=

    2018

  47. [55]

    , title =

    Mathew, Minesh and Karatzas, Dimosthenis and Jawahar, C.V. , title =. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , month =. 2021 , pages =

    2021

  48. [56]

    , title =

    Mathew, Minesh and Bagal, Viraj and Tito, Rub\`en and Karatzas, Dimosthenis and Valveny, Ernest and Jawahar, C.V. , title =. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , month =. 2022 , pages =

    2022

  49. [57]

    Billion-Scale Similarity Search with GPUs , year=

    Johnson, Jeff and Douze, Matthijs and Jégou, Hervé , journal=. Billion-Scale Similarity Search with GPUs , year=

  50. [58]

    The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval , pages =

    Van Gysel, Christophe and de Rijke, Maarten , title =. The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval , pages =. 2018 , isbn =. doi:10.1145/3209978.3210065 , abstract =

    2018

  51. [59]

    2025 , eprint=

    Towards Storage-Efficient Visual Document Retrieval: An Empirical Study on Reducing Patch-Level Embeddings , author=. 2025 , eprint=

    2025

  52. [60]

    2022 , eprint=

    OCR-free Document Understanding Transformer , author=. 2022 , eprint=

    2022

  53. [61]

    2023 , eprint=

    Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding , author=. 2023 , eprint=

    2023

  54. [62]

    2023 , eprint=

    Unifying Vision, Text, and Layout for Universal Document Processing , author=. 2023 , eprint=

    2023

  55. [63]

    2026 , eprint=

    Sculpting the Vector Space: Towards Efficient Multi-Vector Visual Document Retrieval via Prune-then-Merge Framework , author=. 2026 , eprint=

    2026

  56. [64]

    2025 , eprint=

    Hybrid-Vector Retrieval for Visually Rich Documents: Combining Single-Vector Efficiency and Multi-Vector Accuracy , author=. 2025 , eprint=

    2025

  57. [65]

    2605.06460 , archivePrefix=

    Weien Li and Rui Song and Zeyu Li and Haochen Liu and Gonghao Zhang and Difan Jiao and Zhenwei Tang and Bowei He and Haolun Wu and Xue Liu and Ye Yuan , year=. 2605.06460 , archivePrefix=

    arXiv

  58. [66]

    Dao, Tri , booktitle=

  59. [67]

    Advances in neural information processing systems , volume=

    Dao, Tri and Fu, Dan and Ermon, Stefano and Rudra, Atri and R. Advances in neural information processing systems , volume=

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.