Pith. sign in

REVIEW 4 major objections 7 minor 3 cited by

Llama Nemoretriever Colembed: Top-Performing Text-Image Retrieval Model

T0 review · 4 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper introduces llama-nemoretriever-colembed, a unified text-image retrieval model whose 3B variant reports NDCG@5 scores of 91.0 on ViDoRe V1 and 63.5 on ViDoRe V2, placing first on both leaderboards as of June 27, 2025.

desk verdict Strong released models with plausible SOTA scores, but the missing contamination check means the top-1 claim is conditional until the data split question is answered. read the letter →

arxiv 2507.05513 v1 pith:UGY6RNCB submitted 2025-07-07 cs.CV cs.AI

Mengyao Xu , Gabriel Moreira , Ronay Ak , Radek Osmulski , Yauhen Babakhin , Zhiding Yu , Benedikt Schifferer , Even Oldridge This is my paper
classification cs.CVcs.AI
keywords text-imageretrievalvisualdocumentlateinteractionColBERTcontrastivelearningtwo-stagetrainingViDoRebenchmarkmultilingual
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

The paper introduces a family of two text-image retrieval models, llama-nemoretriever-colembed 1B and 3B, and claims the 3B model is the top performer on the ViDoRe V1 and V2 visual document retrieval leaderboards as of June 27, 2025, with NDCG@5 of 91.0 and 63.5. It argues the recipe that gets there is taking the Eagle2 vision-language model, switching causal attention to bidirectional attention, and fine-tuning with a ColBERT-style late-interaction objective in two stages: first on text-only retrieval data, then on text-image data with hard-negative mining. If correct, this shows visual document retrieval can be improved substantially by borrowing text-retrieval pretraining and token-level interaction, not just by scaling the vision model. The paper also quantifies a real cost: late interaction stores thousands of embeddings per image, over 2,700 times more storage than a single-vector bi-encoder, with projection to 512 dimensions reducing that overhead by about 88 percent at a modest accuracy loss.

What carries the argument

The carrying object is the late-interaction bi-encoder built on Eagle2: a shared vision-language encoder that tiles each document image dynamically, outputs one embedding per token, and scores a query by summing per-query-token MaxSim values against stored document token embeddings. That is ColBERT's mechanism transplanted into a vision-language model whose causal attention has been replaced with bidirectional attention. The training machinery is InfoNCE contrastive loss with hard-negative mining (top-k with a 0.95 positive-similarity threshold, K=2), applied first to text-only pairs and then to text-image pairs.

What would settle it

Search the ViDoRe V1 and V2 test queries and documents against the ColPali train set, Wiki-SS-NQ, VDR, VisRAG-Ret-Train, and Docmatix for near-duplicates; if any are found, retrain excluding them and check whether the 91.0 and 63.5 scores and the two first-place rankings survive. Recomputing the leaderboard with the official MTEB Borda-count metric instead of average NDCG@5 would also show whether the 'first' claim depends on the substitute metric.

Watch

Extended reading notes

Core claim

The central claim is that a late-interaction multimodal retriever initialized from the bidirectional-attention Eagle2 backbone and trained with a two-stage contrastive recipe outperforms all published visual document retrieval models on the two ViDoRe benchmarks: NDCG@5 91.0 on V1 and 63.5 on V2, with the 1B variant also beating several 3B and 7B baselines. The paper attributes the gain to two design choices: replacing causal attention with bidirectional attention so every image or text token sees full context, and keeping per-token embeddings so query-document relevance is scored by ColBERT-style MaxSim rather than a single pooled vector. A two-stage schedule first builds text-only retrieval ability on large-scale corpora, then aligns text with images on a multimodal mixture, and hard negatives are mined with a threshold of 95 percent of the positive similarity. The paper's own numbers include the trade-off: the 3B model requires 10,311.1 GB of embeddings per million images at full 3072 dimensions, while projecting to 512 dimensions cuts storage by about 88 percent and reduces ViDoRe V1 from 0.9106 to 0.9064.

Load-bearing premise

The load-bearing premise is that the ViDoRe scores are comparable and untainted: none of the training corpora overlap the test queries or documents, and the reported average NDCG@5 mirrors the official leaderboard ordering.

Editorial extensions

If this is right

  • The 3B model is reported to score 91.0 on ViDoRe V1 and 63.5 on V2, first on both leaderboards as of June 27, 2025, and to hold the top MTEB Visual Document Retrieval score of 83.10.
  • The 2.42B-parameter 1B variant outperforms several 3B and 7B visual retrieval baselines, so a smaller multimodal retriever can beat much larger competitors.
  • Text-only contrastive pretraining transfers to multimodal retrieval, making the two-stage recipe a substantive part of the result rather than a detail.
  • Full-dimensional late interaction costs 10,311.1 GB of stored embeddings per million images for the 3B model; projecting to 512 dimensions cuts storage by about 88 percent with a small NDCG@5 drop.
  • On MIRACL-VISION, the 3B model's mean score of 0.5841 beats prior visual retrieval models on image queries across 18 languages.

Reading between the lines

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

  • The paper does not report an overlap check between its training mixtures (ColPali train set, Wiki-SS-NQ, VDR, VisRAG-Ret-Train, Docmatix) and the ViDoRe test collections, so the reported margins, 91.0 versus 89.9 on V1 and 63.5 versus 60.7 on V2, are not settled against leakage until such a check is done.
  • The 512-dimensional projection's near-parity (0.9106 versus 0.9064 on ViDoRe V1) suggests the effective capacity for late-interaction matching may be far below 3072; testing 256 and 128 dimensions under the same recipe would map the storage-accuracy frontier.
  • The reranker comparison in Table 6 implies the practical production choice is not architectural but budgetary: a bi-encoder with reranking reaches comparable accuracy (0.9064 V1, 0.6214 V2) while storing 3.8 GB per million images instead of thousands of gigabytes.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper introduces llama-nemoretriever-colembed, a family of text-image retrieval models (1B and 3B parameter variants) built on NVIDIA's Eagle2 vision-language model by replacing causal attention with bidirectional attention and adding a ColBERT-style late-interaction scoring. The models are trained in two stages: first text-only contrastive retrieval, then text-image contrastive fine-tuning with hard negative mining. The central empirical claim is that the 3B model achieves NDCG@5 of 91.0 on ViDoRe V1 and 63.5 on ViDoRe V2, ranking first on both leaderboards as of June 27, 2025, and also performs strongly on the MTEB VDR leaderboard and the MIRACL-VISION benchmark. The paper also provides storage and latency trade-off analyses comparing late-interaction and bi-encoder architectures.

Significance. The model release is practically valuable: it provides strong multimodal retrieval performance with competitive parameter efficiency, and the two-stage training recipe with hard negatives is a useful recipe. The storage/latency analysis is a helpful engineering contribution. However, the headline claims are benchmark numbers that hinge on evaluation protocol comparability and on the absence of train/test overlap; neither is fully established, and one of the benchmark claims (MTEB VDR) uses a substituted metric. If the contamination gap is closed and the metric issue clarified, this would be a solid systems contribution.

major comments (4)
  1. [§3.3, §4] The paper does not analyze overlap between the Stage-2 training mixture (ColPali train set, VDR, VisRAG-Ret-Train-In-domain, Docmatix, etc.) and the ViDoRe V1/V2 evaluation sets. Many ViDoRe dataset families (ArxivQA, DocVQA, InfoVQA, TabFQuad, TAT-DQA, and the V2 economics/ESG collections) share source corpora with these training datasets. Because the reported margins over the best baselines are only 1.1 NDCG@5 (V1) and 2.8 NDCG@5 (V2), even a small number of memorized document pages or queries could affect the ranking. Please provide exact and near-duplicate contamination checks at the query and document level, or a clean-split re-evaluation.
  2. [§4, Table 2] The paper substitutes average NDCG@5 for the official MTEB Borda Count, stating that Rank Borda scores are not visible on the MTEB leaderboard, and then claims a 'leading MTEB VDR score of 83.1.' This is not the official MTEB VDR ranking metric, and the phrase could mislead readers into thinking 83.1 is the official score. Please report the official Borda Count values for the models and baselines if available, or explicitly label 83.1 as an unofficial average and avoid calling it the 'MTEB VDR score.' This point is load-bearing because the 'state-of-the-art on MTEB VDR' phrasing appears in the conclusion.
  3. [§4, Table 5] MIRACL-VISION is a benchmark introduced by the same authors (reference [30]), and the paper's claim of 'strong multilingual retrieval capabilities' rests entirely on this self-designed evaluation. The table mixes text-only baselines evaluated on a 'text' subset with visual retrieval models on an 'image' subset, and the set of compared image models is small. Please clarify whether these numbers come from the official MIRACL-VISION evaluation harness, report results for all models with the same protocol, and temper the SOTA claim on this benchmark accordingly.
  4. [§4, §5.2, Table 6] The headline numbers for the 3B model are inconsistent across tables: Table 3 reports ViDoRe V1 91.0 and Table 4 reports V2 63.5, while Table 6 reports 0.9106 and 0.6357 for the same model and notes that a different code base was used. Because the paper itself warns that results 'can slightly change based on the code base,' the exact evaluation code version, commit hash, tiling parameters, and precision must be specified for the numbers in Tables 2–4, and Table 6 should clearly state which protocol it uses to avoid ambiguity about which numbers constitute the SOTA claim.
minor comments (7)
  1. [§4, Table 5] The Table 5 header is corrupted in the manuscript; the column labels are unreadable, making it impossible to verify which models correspond to which scores.
  2. [§5.2, Table 6] Section 5.2 and Table 6 refer to 'lama-3_2-nemoretriever-1b-vlm-embed-v1' (missing the first 'l' in 'llama'); please fix this typo.
  3. [References] Reference [30] lists 'Radek Osmulsk' as an author; this appears to be a typo for 'Radek Osmulski.'
  4. [§4] No error bars or significance tests are reported for any of the NDCG@5 values; given the small margins, a note on variance across query samples would increase confidence.
  5. [§2.1] Section 2.1 states that increasing max_input_tiles from 2 to 4 'showed no performance gains' without providing the supporting experiment; please add the numbers or remove the claim.
  6. [§3.2] The paper does not describe how the pretrained causal-attention weights are initialized or adapted when changing to bidirectional attention; a brief description would improve reproducibility.
  7. [§6] The conclusion claims state-of-the-art results on MIRACL-VISION, but Table 5 includes only a small set of visual retrieval baselines; please either broaden the comparison or qualify the claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central SOTA claim is an external benchmark measurement; no fitted input is renamed as a prediction.

full rationale

The paper's derivation chain is a standard empirical model-building pipeline: initialize from Eagle2, replace causal attention with bidirectional attention, add ColBERT-style MaxSim late interaction, train with InfoNCE and NV-Retriever hard-negative mining, then evaluate on ViDoRe V1/V2 and MIRACL-VISION. The headline numbers (91.0 ViDoRe V1, 63.5 ViDoRe V2) are computed against third-party benchmark corpora and compared with published baselines; they are not produced by the training loss or by any fitted parameter of the model, so there is no equation-to-equation reduction of output to input. The same-author elements (Eagle2 initialization [10,11], NV-Retriever training strategy [2], MIRACL-VISION benchmark [30], reranker trade-off study [31]) are cited as prior work and are externally falsifiable benchmarks or methods rather than unverified premises imported to force the result. The footnote at Section 4 ('The results can slightly change based on the code base... we use Avg NDCG@5 for simplicity') is an evaluation-protocol caveat, and the absence of a train/test overlap analysis between the Stage-2 mixture (ColPali train set, VDR, VisRAG, Docmatix) and ViDoRe dataset families is a contamination or validity risk; both affect correctness, not circularity. Therefore no circular step is identified.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or conceptual entities. Its central claim rests on standard contrastive learning assumptions, a benchmark fairness assumption, and design choices (bidirectional attention, late interaction, hard negative thresholds) that are asserted without ablations.

free parameters (6)
  • InfoNCE temperature tau = not reported
    Temperature in the contrastive loss (Eq. 1); chosen by hand, not stated in the paper.
  • Hard negative threshold = 0.95
    Threshold for 'top-k with percentage to positive' hard negative mining (Section 3.1); selected empirically.
  • Hard negative K = 2
    Number of hard negatives selected per query (Section 3.1); selected empirically.
  • max_input_tiles (training) = 2
    Caps image tiling during training for memory efficiency; increasing to 4 showed no performance gain.
  • max_input_tiles (inference) = 6
    Allows finer visual granularity at inference; chosen by hand.
  • Embedding projection dimension (trade-off variant) = 512
    Linear projection from 3072 to 512 to reduce storage in the Table 6 variant; chosen for the trade-off analysis.
assumptions (4)
  • standard math InfoNCE contrastive loss (Eq. 1) is an appropriate objective for learning retrieval embeddings.
    Uses the standard contrastive learning formulation from reference [18].
  • domain assumption The ViDoRe V1/V2 and MIRACL-VISION benchmarks provide a valid, uncontaminated measure of retrieval quality.
    The paper trains on visual retrieval datasets including the ColPali train set, but no leakage analysis is provided to ensure test sets were not seen.
  • domain assumption MaxSim late interaction (ColBERT-style) is a valid relevance scoring function for multimodal embeddings.
    Adopted from reference [17]; assumes token-level max-similarity sums align with retrieval relevance.
  • ad hoc to paper Replacing causal attention with bidirectional attention is beneficial or at least not harmful for embedding quality.
    Stated in Section 2.1 without an ablation justifying the change.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Llama Nemoretriever Colembed: Top-Performing Text-Image Retrieval Model." pith.science (2026). https://pith.science/paper/UGY6RNCB

@misc{pith2026250705513,
  author       = {Pith},
  title        = {Pith review of: Llama Nemoretriever Colembed: Top-Performing Text-Image Retrieval Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UGY6RNCB}},
  note         = {Machine review of arXiv:2507.05513}
}
read the original abstract

Motivated by the growing demand for retrieval systems that operate across modalities, we introduce llama-nemoretriever-colembed, a unified text-image retrieval model that delivers state-of-the-art performance across multiple benchmarks. We release two model variants, 1B and 3B. The 3B model achieves state of the art performance, scoring NDCG@5 91.0 on ViDoRe V1 and 63.5 on ViDoRe V2, placing first on both leaderboards as of June 27, 2025. Our approach leverages the NVIDIA Eagle2 Vision-Language model (VLM), modifies its architecture by replacing causal attention with bidirectional attention, and integrates a ColBERT-style late interaction mechanism to enable fine-grained multimodal retrieval in a shared embedding space. While this mechanism delivers superior retrieval accuracy, it introduces trade-offs in storage and efficiency. We provide a comprehensive analysis of these trade-offs. Additionally, we adopt a two-stage training strategy to enhance the model's retrieval capabilities.

Figures

Figures reproduced from arXiv: 2507.05513 by the authors.

Figure 1
Figure 1. Multimodal Retrieval Architecture with Dynamic Image Tiling and Late-Interaction Scoring We build our retrieval model on top of the NVIDIA Eagle 2 Vision-Language Models [10, 11]. These models adopt dynamic image tiling to support inputs of varying resolutions, and employ a carefully curated data strategy that improves multimodal learning. These design choices enable Eagle 2 models to achieve state-of-the-art result… view at source ↗
Figure 2
Figure 2. Visualization of bi-encoder and late-interaction architecture 3 Training 3.1 Contrastive learning We leverage contrastive learning to maximize the embedding similarity between the query and positive passage, while minimizing the similarity between the query and negative corpus. We adopt the InfoNCE contrastive loss [18] to train the model to distinguish between positive and negative pairs in a shared embedding space… view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. KoVRE: Training an Efficient Embedding Model for Korean Visual Document Retrieval

    cs.AI 2026-08 conditional novelty 6.0 of 10

    A 2B single-vector retriever, KoVRE, trained with hard-negative mining and reranker distillation, achieves top aggregate Korean visual document retrieval scores.

  2. Col-Bandit: Query-Time Top-$K$ Estimation for Late-Interaction Retrieval

    cs.IR 2026-02 conditional novelty 6.0 of 10

    An adaptive confidence-bound cell-pruning method recovers the exhaustive MaxSim top-K with roughly one-quarter to one-third of the compute on BEIR and REAL-MM-RAG, at the price of a calibrated rather than certified de...

  3. ColGraphRAG: Late-Interaction Evidence Retrieval for Multimodal GraphRAG

    cs.AI 2026-05 conditional novelty 4.0 of 10

    Swapping pooled visual similarity for late-interaction MaxSim in graph-grounded multimodal QA is reported to improve graph-linked image retrieval and QA point estimates on MultimodalQA.

Reference graph

Works this paper leans on

33 extracted references · 14 canonical work pages · cited by 3 Pith papers

  1. [30]

    Miracl-vision: A large, multilingual, visual document retrieval bench- mark.arXiv preprint arXiv:2505.11651, 2025

    Radek Osmulsk, Gabriel de Souza P Moreira, Ronay Ak, Mengyao Xu, Benedikt Schifferer, and Even Oldridge. Miracl-vision: A large, multilingual, visual document retrieval bench- mark.arXiv preprint arXiv:2505.11651, 2025

    arXiv 2025

  2. [1]

    Nv-embed: Improved techniques for training llms as generalist embedding models.arXiv preprint arXiv:2405.17428, 2024

    Chankyu Lee, Rajarshi Roy, Mengyao Xu, Jonathan Raiman, Mohammad Shoeybi, Bryan Catanzaro, and Wei Ping. Nv-embed: Improved techniques for training llms as generalist embedding models.arXiv preprint arXiv:2405.17428, 2024. 10Core Contributor 9 Model Avg. ESGRestaurantHuman EconomicsMacroMultilingual MITBiomedical ESGRestaurantSynthetic ESG RestaurantSynth...

    arXiv 2024

  3. [2]

    Nv-retriever: Improving text embedding models with effective hard- negative mining.arXiv preprint arXiv:2407.15831, 2024

    Gabriel de Souza P Moreira, Radek Osmulski, Mengyao Xu, Ronay Ak, Benedikt Schifferer, and Even Oldridge. Nv-retriever: Improving text embedding models with effective hard- negative mining.arXiv preprint arXiv:2407.15831, 2024

    arXiv 2024

  4. [3]

    Qwen3 embed- ding: Advancing text embedding and reranking through foundation models.arXiv preprint arXiv:2506.05176, 2025

    Y anzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Y ang, Pengjun Xie, An Y ang, Dayiheng Liu, Junyang Lin, Fei Huang, and Jingren Zhou. Qwen3 embed- ding: Advancing text embedding and reranking through foundation models.arXiv preprint arXiv:2506.05176, 2025

    arXiv 2025

  5. [4]

    Improving text embeddings with large language models.arXiv preprint arXiv:2401.00368, 2023

    Liang Wang, Nan Y ang, Xiaolong Huang, Linjun Y ang, Rangan Majumder, and Furu Wei. Improving text embeddings with large language models.arXiv preprint arXiv:2401.00368, 2023

    arXiv 2023

  6. [5]

    Mteb: Massive text embedding benchmark.arXiv preprint arXiv:2210.07316, 2022

    Niklas Muennighoff, Nouamane Tazi, Loïc Magne, and Nils Reimers. Mteb: Massive text embedding benchmark.arXiv preprint arXiv:2210.07316, 2022

    arXiv 2022

  7. [6]

    Maintaining mteb: Towards long term usability and reproducibility of embedding bench- marks.arXiv preprint arXiv:2506.21182, 2025

    Isaac Chung, Imene Kerboua, Marton Kardos, Roman Solomatin, and Kenneth Enevoldsen. Maintaining mteb: Towards long term usability and reproducibility of embedding bench- marks.arXiv preprint arXiv:2506.21182, 2025

    arXiv 2025

  8. [7]

    Colpali: Efficient document retrieval with vision language models, 2024

    Manuel Faysse, Hugues Sibille, Tony Wu, Bilel Omrani, Gautier Viaud, Céline Hudelot, and Pierre Colombo. Colpali: Efficient document retrieval with vision language models, 2024

    work page 2024

Show all 33 references
  1. [8]

    Qwen2-vl: Enhancing vision-language model’s percep- tion of the world at any resolution.arXiv preprint arXiv:2409.12191, 2024

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, et al. Qwen2-vl: Enhancing vision-language model’s percep- tion of the world at any resolution.arXiv preprint arXiv:2409.12191, 2024

    2024 arXiv

  2. [9]

    Llama-nemotron: Efficient reasoning models.arXiv preprint arXiv:2505.00949, 2025

    Akhiad Bercovich, Itay Levy, Izik Golan, Mohammad Dabbah, Ran El-Y aniv, Omri Puny, Ido Galil, Zach Moshe, Tomer Ronen, Najeeb Nabwani, et al. Llama-nemotron: Efficient reasoning models.arXiv preprint arXiv:2505.00949, 2025

    2025

  3. [10]

    Eagle 2: Building post-training data strategies from scratch for frontier vision-language models.arXiv preprint arXiv:2501.14818, 2025

    Zhiqi Li, Guo Chen, Shilong Liu, Shihao Wang, Vibashan VS, Yishen Ji, Shiyi Lan, Hao Zhang, Yilin Zhao, Subhashree Radhakrishnan, et al. Eagle 2: Building post-training data strategies from scratch for frontier vision-language models.arXiv preprint arXiv:2501.14818, 2025

    2025 arXiv

  4. [11]

    Eagle 2.5: Boosting long-context post-training for frontier vision-language models.arXiv preprint arXiv:2504.15271, 2025

    Guo Chen, Zhiqi Li, Shihao Wang, Jindong Jiang, Yicheng Liu, Lidong Lu, De-An Huang, Wonmin Byeon, Matthieu Le, Tuomas Rintamaki, et al. Eagle 2.5: Boosting long-context post-training for frontier vision-language models.arXiv preprint arXiv:2504.15271, 2025. 10

    2025

  5. [12]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agar- wal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. InInternational conference on machine learning, p...

    2021

  6. [13]

    Siglip 2: Multilingual vision-language encoders with improved semantic understanding, lo- calization, and dense features.arXiv preprint arXiv:2502.14786, 2025

    Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muhammad Ferjad Naeem, Ibrahim Al- abdulmohsin, Nikhil Parthasarathy, Talfan Evans, Lucas Beyer, Y e Xia, Basil Mustafa, et al. Siglip 2: Multilingual vision-language encoders with improved semantic understanding, lo- calization,...

    2025 arXiv

  7. [14]

    Am-radio: Agglomera- tive vision foundation model reduce all domains into one

    Mike Ranzinger, Greg Heinrich, Jan Kautz, and Pavlo Molchanov. Am-radio: Agglomera- tive vision foundation model reduce all domains into one. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12490–12500, 2024

    2024

  8. [15]

    Vidore benchmark v2: Raising the bar for visual retrieval.arXiv preprint arXiv:2505.17166, 2025

    Quentin Macé, António Loison, and Manuel Faysse. Vidore benchmark v2: Raising the bar for visual retrieval.arXiv preprint arXiv:2505.17166, 2025

    2025

  9. [16]

    Text embeddings by weakly-supervised contrastive pre-training

    Liang Wang, Nan Y ang, Xiaolong Huang, Binxing Jiao, Linjun Y ang, Daxin Jiang, Rangan Majumder, and Furu Wei. Text embeddings by weakly-supervised contrastive pre-training. arXiv preprint arXiv:2212.03533, 2022

    2022 arXiv

  10. [17]

    Colbert: Efficient and effective passage search via con- textualized late interaction over bert

    Omar Khattab and Matei Zaharia. Colbert: Efficient and effective passage search via con- textualized late interaction over bert. InProceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval, pages 39–48, 2020

    2020

  11. [18]

    A simple frame- work for contrastive learning of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple frame- work for contrastive learning of visual representations. InInternational conference on ma- chine learning, pages 1597–1607. PmLR, 2020

    2020

  12. [19]

    Hotpotqa: A dataset for diverse, explainable multi-hop question answering.arXiv preprint arXiv:1809.09600, 2018

    Zhilin Y ang, Peng Qi, Saizheng Zhang, Y oshua Bengio, William W Cohen, Ruslan Salakhut- dinov, and Christopher D Manning. Hotpotqa: A dataset for diverse, explainable multi-hop question answering.arXiv preprint arXiv:1809.09600, 2018

    2018 arXiv

  13. [20]

    Miracl: A mul- tilingual retrieval dataset covering 18 diverse languages.Transactions of the Association for Computational Linguistics, 11:1114–1131, 2023

    Xinyu Zhang, Nandan Thakur, Odunayo Ogundepo, Ehsan Kamalloo, David Alfonso- Hermelo, Xiaoguang Li, Qun Liu, Mehdi Rezagholizadeh, and Jimmy Lin. Miracl: A mul- tilingual retrieval dataset covering 18 diverse languages.Transactions of the Association for Computational Linguist...

    2023

  14. [21]

    Natural questions: a benchmark for question answering research.Transactions of the Association for Computational Linguistics, 7:453–466, 2019

    Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, et al. Natural questions: a benchmark for question answering research.Transactions of the Association for Computati...

    2019

  15. [22]

    Stack Exchange Community Data Dump, 2023.https://archive

    Stack Exchange, Inc. Stack Exchange Community Data Dump, 2023.https://archive. org/details/stack-exchange-data-dump-2023-09-12, 2023. Accessed: 2025-06-30

    2023

  16. [23]

    Squad: 100,000+ ques- tions for machine comprehension of text.arXiv preprint arXiv:1606.05250, 2016

    Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. Squad: 100,000+ ques- tions for machine comprehension of text.arXiv preprint arXiv:1606.05250, 2016

    2016 arXiv

  17. [24]

    Mammoth2: Scaling instructions from the web.Advances in Neural Information Processing Systems, 2024

    Xiang Y ue, Tuney Zheng, Ge Zhang, and Wenhu Chen. Mammoth2: Scaling instructions from the web.Advances in Neural Information Processing Systems, 2024

    2024

  18. [25]

    Xueguang Ma

    et al. Xueguang Ma. Tevatron/wiki-ss-nq, 2024. 11

    2024

  19. [26]

    Tevatron 2.0: Unified document retrieval toolkit across scale, language, and modality

    Xueguang Ma, Luyu Gao, Shengyao Zhuang, Jiaqi Samantha Zhan, Jamie Callan, and Jimmy Lin. Tevatron 2.0: Unified document retrieval toolkit across scale, language, and modality. arXiv preprint arXiv:2505.02466, 2025

    2025 arXiv

  20. [27]

    llamaindex/vdr-multilingual-train, 2025

    Marco Cimolai and Logan Markewich. llamaindex/vdr-multilingual-train, 2025

    2025

  21. [28]

    Visrag: Vision-based retrieval-augmented generation on multi-modality documents, 2024

    Shi Y u, Chaoyue Tang, Bokai Xu, Junbo Cui, Junhao Ran, Y ukun Y an, Zhenghao Liu, Shuo Wang, Xu Han, Zhiyuan Liu, and Maosong Sun. Visrag: Vision-based retrieval-augmented generation on multi-modality documents, 2024

    2024

  22. [29]

    Building and better understanding vision-language models: insights and future directions., 2024

    Hugo Laurençon, Andrés Marafioti, Victor Sanh, and Léo Tronchon. Building and better understanding vision-language models: insights and future directions., 2024

    2024

  23. [31]

    Enhancing q&a text retrieval with ranking models: Benchmarking, fine- tuning and deploying rerankers for rag

    Gabriel de Souza P Moreira, Ronay Ak, Benedikt Schifferer, Mengyao Xu, Radek Osmulski, and Even Oldridge. Enhancing q&a text retrieval with ranking models: Benchmarking, fine- tuning and deploying rerankers for rag. InProceedings of the 1st Workshop on GenAI and RAG Systems for...

    2024

  24. [32]

    A little pooling goes a long way for multi-vector representations, 2024

    Benjamin Clavié. A little pooling goes a long way for multi-vector representations, 2024

    2024

  25. [33]

    Muvera: multi-vector retrieval via fixed dimensional encodings.arXiv preprint arXiv:2405.19504, 2024

    Laxman Dhulipala, Majid Hadian, Rajesh Jayaram, Jason Lee, and V ahab Mirrokni. Muvera: multi-vector retrieval via fixed dimensional encodings.arXiv preprint arXiv:2405.19504, 2024. 12

    2024 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.