The reviewed record of science sign in
Pith

arxiv: 2606.19960 · v1 · pith:ES6JX3RT · submitted 2026-06-18 · cs.IR

Stellar: Scalable Multimodal Document Retrieval for Natural Language Queries

Yuxiang Guo , Zhonghao Hu , Yuren Mao , Yuhang Liu , Congcong Ge , Xiaolu Zhang , Jun Zhou , Yunjun Gao This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-26 15:48 UTCgrok-4.3pith:ES6JX3RTrecord.jsonopen to challenge →

classification cs.IR
keywords multimodal document retrievallate interactionlexical filteringdisk-backed storagescalable retrievalMLLMRAG systemstoken embeddings
0
0 comments X

The pith

Stellar reduces memory overhead and query latency for multimodal document retrieval by 1-2 orders of magnitude without losing effectiveness.

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

The paper shows how to make multi-vector retrieval scalable by keeping most token embeddings on disk and using a fine-tuned multimodal model to filter down to a small candidate set before late interaction. This addresses the high memory cost that has limited deployment of accurate retrieval methods in RAG systems. If the approach works, large-scale multimodal search becomes feasible on ordinary hardware while preserving the quality of current state-of-the-art methods. Experiments on four benchmarks plus a new large dataset back the claim.

Core claim

Stellar introduces Lexical Representation-based Filtering that fine-tunes an MLLM as a sparse encoder for effective document pruning, paired with Efficient Disk-backed Late Interaction that uses balanced clustering to lay out token embeddings on disk and a cost model to load only needed ones into memory for late interaction.

What carries the argument

Lexical Representation-based Filtering (LRF) using a fine-tuned MLLM as sparse encoder combined with Efficient Disk-backed Late Interaction (DLI) using balanced clustering for on-disk storage.

If this is right

  • Multimodal document retrieval can now operate on corpora much larger than current memory limits allow.
  • Production RAG systems can adopt late-interaction methods without requiring specialized high-RAM servers.
  • Query response times improve enough to support interactive applications at scale.
  • The same framework may extend to other embedding-based retrieval tasks that suffer from multi-vector memory costs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Success would shift design focus from in-memory indexes to hybrid disk-memory architectures for retrieval.
  • Further gains might come from optimizing the clustering or cost model for specific hardware.
  • Testing on even larger synthetic datasets could reveal where the filtering step starts to drop recall.

Load-bearing premise

The fine-tuned MLLM used for lexical filtering will not exclude any truly relevant documents from the small candidate set passed to late interaction.

What would settle it

Running the LRF step on a benchmark and finding that for some queries the gold relevant document is not in the filtered candidate set.

Figures

Figures reproduced from arXiv: 2606.19960 by Congcong Ge, Jun Zhou, Xiaolu Zhang, Yuhang Liu, Yunjun Gao, Yuren Mao, Yuxiang Guo, Zhonghao Hu.

Figure 1
Figure 1. Figure 1: Multimodal document retrieval for NL queries. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: An illustration of the single-vector-based method and multi-vector [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of STELLAR. (a) Lexical Representation-based Filtering enables efficient and effective document filtering. (b) Efficient Disk-backed Late Interaction supports low-cost multi-vector similarity computation and sparse-dense score fusion. C. Late Interaction Mechanism To capture fine-grained semantic alignments between queries and documents, multi-vector representation learn￾ing [14] avoids compressin… view at source ↗
Figure 4
Figure 4. Figure 4: Runtime of lexical representation-based filtering (LRF) and disk [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Average number of non-zero dimensions of our sparse representations [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 8
Figure 8. Figure 8: Sensitivity analysis of α in score fusion. Filter, which filters candidates using centroids obtained by clustering token-level embeddings. We show Recall@k1 across different k1 on a small-scale dataset and a large-scale dataset in [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
read the original abstract

Multimodal document retrieval--selecting the most relevant multimodal document from a large corpus to answer a natural language query--plays an essential role in Retrieval-Augmented Generation (RAG) systems. State-of-the-art methods represent each document and query with multiple token-level embeddings and use late interaction to achieve high effectiveness. However, such multi-vector representations incur substantial memory overhead during retrieval, leading to poor scalability and hindering real-world deployment. In this paper, we present Stellar, a scalable multimodal document retrieval framework that stores token-level document embeddings on disk and loads only a small set of candidate embeddings into memory for late interaction. Stellar comprises two key components: (i) Lexical Representation-based Filtering (LRF), which fine-tunes a Multimodal Large Language Model (MLLM) as a sparse encoder to produce high-quality lexical representations, enabling efficient and effective document filtering to significantly reduce the candidate set; (ii) Efficient Disk-backed Late Interaction (DLI), which designs an on-disk token embedding storage layout guided by a balanced clustering algorithm, and dynamically loads only the necessary token embeddings into memory using a simple yet effective cost model. Extensive experiments on four real-world benchmarks and a newly presented large-scale dataset demonstrate that Stellar reduces memory overhead and query latency by 1-2 orders of magnitude compared to existing methods without compromising retrieval effectiveness.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces Stellar, a multimodal document retrieval framework for natural language queries that combines Lexical Representation-based Filtering (LRF) — fine-tuning an MLLM as a sparse encoder for candidate reduction — with Efficient Disk-backed Late Interaction (DLI) that stores token embeddings on disk and loads only necessary candidates via a balanced clustering layout and cost model. It claims 1-2 orders of magnitude reductions in memory and query latency versus prior late-interaction methods on four benchmarks plus a new large-scale dataset, while preserving retrieval effectiveness.

Significance. If the central effectiveness claim holds, Stellar would materially improve the practicality of token-level late-interaction retrieval in RAG pipelines by addressing the memory and latency bottlenecks that currently limit deployment at scale. The introduction of a new large-scale multimodal dataset is a concrete contribution that could support future work.

major comments (2)
  1. [§3.2 and §4] §3.2 (LRF) and §4 (experiments): the headline claim that effectiveness is uncompromised rests on the assumption that LRF produces a candidate set with sufficiently high recall that the subsequent DLI stage matches full-corpus late-interaction performance. The manuscript reports only end-to-end metrics; the critical intermediate recall@K (or equivalent) for the LRF sparse-encoder stage alone is not provided, leaving the load-bearing assumption unverified.
  2. [§4] §4 (experimental setup): the abstract and results claim reductions “without compromising retrieval effectiveness,” yet no details are given on the precise baselines, statistical significance testing, or whether any post-hoc threshold tuning was performed on the LRF candidate-set size; this information is required to assess whether the reported effectiveness numbers are robust.
minor comments (2)
  1. [Abstract] The abstract states positive outcomes on “four real-world benchmarks and a newly presented large-scale dataset” but provides no metric names, dataset sizes, or baseline references; adding one sentence with these specifics would improve readability.
  2. [§3.3] Notation for the cost model in DLI is introduced without an explicit equation number; cross-referencing the model to a numbered equation would aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback on our manuscript. We address each of the major comments below and outline the revisions we will make.

read point-by-point responses

  1. Referee: [§3.2 and §4] §3.2 (LRF) and §4 (experiments): the headline claim that effectiveness is uncompromised rests on the assumption that LRF produces a candidate set with sufficiently high recall that the subsequent DLI stage matches full-corpus late-interaction performance. The manuscript reports only end-to-end metrics; the critical intermediate recall@K (or equivalent) for the LRF sparse-encoder stage alone is not provided, leaving the load-bearing assumption unverified.

    Authors: We agree that the intermediate recall of the LRF stage is important to report for full transparency. While the end-to-end results demonstrate that Stellar achieves comparable effectiveness to full late-interaction methods, adding the LRF recall metrics will better support our claims. We will include these results in the revised manuscript, showing high recall at the chosen candidate sizes. revision: yes

  2. Referee: [§4] §4 (experimental setup): the abstract and results claim reductions “without compromising retrieval effectiveness,” yet no details are given on the precise baselines, statistical significance testing, or whether any post-hoc threshold tuning was performed on the LRF candidate-set size; this information is required to assess whether the reported effectiveness numbers are robust.

    Authors: We will provide additional details in the experimental setup section, including the exact list of baselines compared, results of statistical significance tests (e.g., p-values from t-tests), and clarification that the candidate set size was selected based on validation data without post-hoc tuning on the test sets. This will address concerns about robustness. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical architecture validated by benchmarks

full rationale

The paper describes a retrieval system (LRF sparse filtering via fine-tuned MLLM plus disk-backed DLI) whose central claims are measured end-to-end effectiveness, memory, and latency on four benchmarks plus a new dataset. No equations, uniqueness theorems, or predictions are presented that reduce by construction to fitted inputs, self-citations, or renamed ansatzes. All load-bearing steps are external measurements rather than internal redefinitions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract describes an engineering system without explicit mathematical derivations or new theoretical entities; relies on standard assumptions of IR evaluation and MLLM fine-tuning effectiveness.

pith-pipeline@v0.9.1-grok · 5788 in / 1049 out tokens · 26382 ms · 2026-06-26T15:48:03.610870+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 7 canonical work pages · 4 internal anchors

  1. [1]

    Language models are few-shot learners,

    T. B. Brown, B. Mann, N. Ryder, M. Subbiah, J. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, S. Agarwal, A. Herbert- V osset al., “Language models are few-shot learners,” inNeurIPS, 2020

    2020

  2. [2]

    GPT-4o System Card

    A. Hurst, A. Lerer, A. P. Goucher, A. Perelman, A. Ramesh, A. Clark, A. Ostrow, A. Welihinda, A. Hayes, A. Radfordet al., “Gpt-4o system card,”arXiv preprint arXiv:2410.21276, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  3. [3]

    A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions,

    L. Huang, W. Yu, W. Ma, W. Zhong, Z. Feng, H. Wang, Q. Chen, W. Peng, X. Feng, B. Qin, and T. Liu, “A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions,” ACM Trans. Inf. Syst., vol. 43, no. 2, pp. 42:1–42:55, 2025

    2025

  4. [4]

    A survey on RAG meeting llms: Towards retrieval-augmented large language models,

    W. Fan, Y . Ding, L. Ning, S. Wang, H. Li, D. Yin, T. Chua, and Q. Li, “A survey on RAG meeting llms: Towards retrieval-augmented large language models,” inSIGKDD, 2024, pp. 6491–6501

    2024

  5. [5]

    Self-rag: Learning to retrieve, generate, and critique through self-reflection,

    A. Asai, Z. Wu, Y . Wang, A. Sil, and H. Hajishirzi, “Self-rag: Learning to retrieve, generate, and critique through self-reflection,” inICLR, 2024

    2024

  6. [6]

    FIT-RAG: black-box RAG with factual information and token reduction,

    Y . Mao, X. Dong, W. Xu, Y . Gao, B. Wei, and Y . Zhang, “FIT-RAG: black-box RAG with factual information and token reduction,”ACM Trans. Inf. Syst., vol. 43, no. 2, pp. 40:1–40:27, 2025

    2025

  7. [7]

    QAEA-DR: A unified text augmentation framework for dense retrieval,

    H. Tan, S. Zhan, H. Lin, H. Zheng, and W. K. Chan, “QAEA-DR: A unified text augmentation framework for dense retrieval,”IEEE Trans. Knowl. Data Eng., vol. 37, no. 6, pp. 3669–3683, 2025

    2025

  8. [8]

    Docdb: A database for unstructured document analysis,

    Z. Li, Y . Zhong, C. Chai, Z. Sun, Y . Deng, Y . Yuan, G. Wang, and L. Cao, “Docdb: A database for unstructured document analysis,”Proc. VLDB Endow., vol. 18, no. 12, pp. 5387–5390, 2025

    2025

  9. [9]

    Visrag: Vision-based retrieval-augmented generation on multi-modality documents,

    S. Yu, C. Tang, B. Xu, J. Cui, J. Ran, Y . Yan, Z. Liu, S. Wang, X. Han, Z. Liu, and M. Sun, “Visrag: Vision-based retrieval-augmented generation on multi-modality documents,” inICLR, 2025

    2025

  10. [10]

    Robust scene text recognition with automatic rectification,

    B. Shi, X. Wang, P. Lyu, C. Yao, and X. Bai, “Robust scene text recognition with automatic rectification,” inCVPR, 2016, pp. 4168– 4176

    2016

  11. [11]

    The probabilistic relevance frame- work: BM25 and beyond,

    S. E. Robertson and H. Zaragoza, “The probabilistic relevance frame- work: BM25 and beyond,”Found. Trends Inf. Retr., vol. 3, no. 4, pp. 333–389, 2009

    2009

  12. [12]

    C-Pack: Packed Resources For General Chinese Embeddings

    S. Xiao, Z. Liu, P. Zhang, and N. Muennighoff, “C-pack: Pack- aged resources to advance general chinese embedding,”CoRR, vol. abs/2309.07597, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  13. [13]

    A Survey on Multimodal Large Language Models

    S. Yin, C. Fu, S. Zhao, K. Li, X. Sun, T. Xu, and E. Chen, “A survey on multimodal large language models,”arXiv preprint arXiv:2306.13549, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  14. [14]

    Colpali: Efficient document retrieval with vision language models,

    M. Faysse, H. Sibille, T. Wu, B. Omrani, G. Viaud, C. Hudelot, and P. Colombo, “Colpali: Efficient document retrieval with vision language models,” inICLR, 2025

    2025

  15. [15]

    Unifying multimodal retrieval via document screenshot embedding,

    X. Ma, S. Lin, M. Li, W. Chen, and J. Lin, “Unifying multimodal retrieval via document screenshot embedding,” inEMNLP, 2024, pp. 6492–6505

    2024

  16. [16]

    2411.04952 , archivePrefix=

    J. Cho, D. Mahata, O. Irsoy, Y . He, and M. Bansal, “M3docrag: Multi-modal retrieval is what you need for multi-page multi-document understanding,”arXiv preprint arXiv:2411.04952, 2024

    work page arXiv 2024

  17. [17]

    Colbert: Efficient and effective passage search via contextualized late interaction over BERT,

    O. Khattab and M. Zaharia, “Colbert: Efficient and effective passage search via contextualized late interaction over BERT,” inSIGIR, 2020, pp. 39–48

    2020

  18. [18]

    Building and better understanding vision-language models: insights and future directions

    H. Laurenc ¸on, A. Marafioti, V . Sanh, and L. Tronchon, “Building and better understanding vision-language models: insights and future directions.” inarXiv preprint arXiv:2408.12637, 2024

    work page arXiv 2024

  19. [19]

    PLAID: an efficient engine for late interaction retrieval,

    K. Santhanam, O. Khattab, C. Potts, and M. Zaharia, “PLAID: an efficient engine for late interaction retrieval,” inCIKM, 2022, pp. 1747– 1756

    2022

  20. [20]

    Colbertv2: Effective and efficient retrieval via lightweight late interac- tion,

    K. Santhanam, O. Khattab, J. Saad-Falcon, C. Potts, and M. Zaharia, “Colbertv2: Effective and efficient retrieval via lightweight late interac- tion,” inNAACL-HLT, 2022, pp. 3715–3734

    2022

  21. [21]

    Effective multi-modal retrieval based on stacked auto-encoders,

    W. Wang, B. C. Ooi, X. Yang, D. Zhang, and Y . Zhuang, “Effective multi-modal retrieval based on stacked auto-encoders,”Proc. VLDB Endow., vol. 7, no. 8, pp. 649–660, 2014

    2014

  22. [22]

    Chameleon: Foundation models for fairness-aware multi-modal data augmentation to enhance coverage of minorities,

    M. Erfanian, H. V . Jagadish, and A. Asudeh, “Chameleon: Foundation models for fairness-aware multi-modal data augmentation to enhance coverage of minorities,”Proc. VLDB Endow., vol. 17, no. 11, pp. 3470– 3483, 2024

    2024

  23. [23]

    Alleviating the inconsistency of multimodal data in cross-modal retrieval,

    T. Li, X. Yang, Y . Ke, B. Wang, Y . Liu, and J. Xu, “Alleviating the inconsistency of multimodal data in cross-modal retrieval,” inICDE, 2024, pp. 4643–4656

    2024

  24. [24]

    MUST: an effective and scalable framework for multimodal search of target modality,

    M. Wang, X. Ke, X. Xu, L. Chen, Y . Gao, P. Huang, and R. Zhu, “MUST: an effective and scalable framework for multimodal search of target modality,” inICDE, 2024, pp. 4747–4759

    2024

  25. [25]

    Learning transferable visual models from natural language supervi- sion,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning transferable visual models from natural language supervi- sion,” inICML, 2021, pp. 8748–8763

    2021

  26. [26]

    BLIP-2: bootstrapping language-image pre-training with frozen image encoders and large language models,

    J. Li, D. Li, S. Savarese, and S. C. H. Hoi, “BLIP-2: bootstrapping language-image pre-training with frozen image encoders and large language models,” inICML, 2023, pp. 19 730–19 742

    2023

  27. [27]

    Sigmoid loss for language image pre-training,

    X. Zhai, B. Mustafa, A. Kolesnikov, and L. Beyer, “Sigmoid loss for language image pre-training,” inICCV, 2023, pp. 11 941–11 952

    2023

  28. [28]

    Momentum contrast for unsupervised visual representation learning,

    K. He, H. Fan, Y . Wu, S. Xie, and R. Girshick, “Momentum contrast for unsupervised visual representation learning,” inCVPR, 2020, pp. 9729–9738

    2020

  29. [29]

    Efficient and robust approxi- mate nearest neighbor search using hierarchical navigable small world graphs,

    Y . A. Malkov and D. A. Yashunin, “Efficient and robust approxi- mate nearest neighbor search using hierarchical navigable small world graphs,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 42, no. 4, pp. 824–836, 2018

    2018

  30. [30]

    Diskann: fast accurate billion-point nearest neighbor search on a single node,

    S. J. Subramanya, Devvrit, R. Kadekodi, R. Krishaswamy, and H. V . Simhadri, “Diskann: fast accurate billion-point nearest neighbor search on a single node,” inNeurIPS, 2019, pp. 13 766–13 776

    2019

  31. [31]

    Starling: An i/o-efficient disk-resident graph index framework for high-dimensional vector similarity search on data segment,

    M. Wang, W. Xu, X. Yi, S. Wu, Z. Peng, X. Ke, Y . Gao, X. Xu, R. Guo, and C. Xie, “Starling: An i/o-efficient disk-resident graph index framework for high-dimensional vector similarity search on data segment,”Proc. ACM Manag. Data, vol. 2, no. 1, pp. 14:1–14:27, 2024

    2024

  32. [32]

    Milvus: A purpose-built vector data management system,

    J. Wang, X. Yi, R. Guo, H. Jin, P. Xu, S. Li, X. Wang, X. Guo, C. Li, X. Xu, K. Yu, Y . Yuan, Y . Zou, J. Long, Y . Cai, Z. Li, Z. Zhang, Y . Mo, J. Gu, R. Jiang, Y . Wei, and C. Xie, “Milvus: A purpose-built vector data management system,” inSIGMOD, 2021, pp. 2614–2627

    2021

  33. [33]

    Splade v2: Sparse lexi- cal and expansion model for first stage retrieval,

    M. Formal, X. Martinet, N. Thakur, H. Sajjad, C. Lioma, F. Scholer, M. Worring, M. Bendersky, and F. Diaz, “Splade v2: Sparse lexi- cal and expansion model for first stage retrieval,” inarXiv preprint arXiv:2106.05222, 2021

    work page arXiv 2021

  34. [34]

    Bert: Pre-training of deep bidirectional transformers for language understanding,

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,” inNAACL- HLT, 2019, pp. 4171–4186

    2019

  35. [35]

    Stair: Learning sparse text and image representation in grounded tokens,

    C. Chen, B. Zhang, L. Cao, J. Shen, T. Gunter, A. M. Jose, A. Toshev, Y . Zheng, J. Shlens, R. Pang, and Y . Yang, “Stair: Learning sparse text and image representation in grounded tokens,” inEMNLP, 2023, pp. 15 079–15 094

    2023

  36. [36]

    Multimodal learned sparse retrieval with probabilistic expansion control,

    T. Nguyen, M. Hendriksen, A. Yates, and M. de Rijke, “Multimodal learned sparse retrieval with probabilistic expansion control,” inECIR, 2024, pp. 114–131

    2024

  37. [37]

    BIRDIE: nat- ural language-driven table discovery using differentiable search index,

    Y . Guo, Z. Hu, Y . Mao, B. Zheng, Y . Gao, and M. Zhou, “BIRDIE: nat- ural language-driven table discovery using differentiable search index,” Proc. VLDB Endow., vol. 18, no. 7, pp. 2070–2083, 2025

    2070

  38. [38]

    Lora: Low-rank adaptation of large language models,

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, and W. Chen, “Lora: Low-rank adaptation of large language models,” inICLR, 2022

    2022

  39. [39]

    I/o-efficient algorithms and environments,

    J. S. Vitteret al., “I/o-efficient algorithms and environments,”ACM Computing Surveys, vol. 31, no. 2, pp. 256–257, 1999

    1999

  40. [40]

    Pneuma:leveraging llms for tabular data representation and retrieval in an end-to-end system,

    M. I. L. Balaka, D. Alexander, Q. Wang, Y . Gong, A. Krisnadhi, and R. Castro Fernandez, “Pneuma:leveraging llms for tabular data representation and retrieval in an end-to-end system,”Proc. ACM Manag. Data, vol. 3, no. 3, 2025

    2025

  41. [41]

    Hierarchical multimodal trans- formers for multipage docvqa,

    R. Tito, D. Karatzas, and E. Valveny, “Hierarchical multimodal trans- formers for multipage docvqa,”Pattern Recognit., vol. 144, p. 109834, 2023

    2023

  42. [42]

    Infographicvqa,

    M. Mathew, V . Bagal, R. Tito, D. Karatzas, E. Valveny, and C. V . Jawahar, “Infographicvqa,” inWACV, 2022, pp. 2582–2591

    2022

  43. [43]

    Multimodal arxiv: A dataset for improving scientific comprehension of large vision-language models,

    L. Li, Y . Wang, R. Xu, P. Wang, X. Feng, L. Kong, and Q. Liu, “Multimodal arxiv: A dataset for improving scientific comprehension of large vision-language models,” inACL, 2024, pp. 14 369–14 387

    2024

  44. [44]

    Plotqa: Rea- soning over scientific plots,

    N. Methani, P. Ganguly, M. M. Khapra, and P. Kumar, “Plotqa: Rea- soning over scientific plots,” inWACV, 2020, pp. 1527–1536

    2020

  45. [45]

    Paddleocr: Multi-language ocr toolkit based on paddlepaddle,

    “Paddleocr: Multi-language ocr toolkit based on paddlepaddle,” https: //github.com/PaddlePaddle/PaddleOCR, 2020

    2020

  46. [46]

    Nv-embed: Improved techniques for training llms as generalist embedding models,

    C. Lee, R. Roy, M. Xu, J. Raiman, M. Shoeybi, B. Catanzaro, and W. Ping, “Nv-embed: Improved techniques for training llms as generalist embedding models,” inICLR, 2025

    2025

  47. [47]

    Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution

    P. Wang, S. Bai, S. Tan, S. Wang, Z. Fan, J. Bai, K. Chen, X. Liu, J. Wang, W. Ge, Y . Fan, K. Dang, M. Du, X. Ren, R. Men, D. Liu, C. Zhou, J. Zhou, and J. Lin, “Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution,”arXiv preprint arXiv:2409.12191, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024