Pith. sign in

REVIEW 4 major objections 5 minor 31 references

Self-Consistency in Vision-Language Models for Precision Agriculture: Multi-Response Consensus for Crop Disease Management

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a multi-response consensus vote over fine-tuned embeddings raises a vision-language model's maize leaf blight diagnosis accuracy from 82.2% to 87.8%, and lifts treatment recommendation accuracy from 27.8% to 43.3%.

desk verdict A plausible, clearly-written engineering contribution whose headline accuracy gains are measured by a circular o1-mini scorer that also created the ground truth and trained the embedding; the numbers are not yet established. read the letter →

arxiv 2507.08024 v1 pith:X5432EQT submitted 2025-07-08 cs.CV

Mihir Gupta , Abhay Mangla , Ross Greer , Pratik Desai This is my paper · ORCID
classification cs.CV
keywords precisionagriculturevision-languagemodelscropdiseasedetectionself-consistencyvotingmaizeleafblightembeddingfine-tuningmobiledeploymentprompt-basedevaluation
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

Applied to maize leaf blight diagnosis from farmer photos, this paper claims that a vision-language model becomes more reliable when it generates several candidate answers and then votes among them using an embedding space trained on agricultural expertise. The proposed pipeline raises disease identification accuracy from 82.2% to 87.8%, symptom analysis from 38.9% to 52.2%, and treatment recommendation from 27.8% to 43.3% relative to standard single-answer decoding on the same fine-tuned PaliGemma model. The point that would matter for precision agriculture is that these gains come without retraining the generative model or calling a large language model at inference, so the improvement fits inside a lightweight mobile deployment.

What carries the argument

The argument is carried by two linked components. The first is a prompt-based scorer: the o1-mini language model is instructed as an expert plant pathologist to compare responses point-by-point for disease identity, symptom severity, and treatment equivalence, returning a 0–1 similarity score; the same scorer produces the training labels for the embedding model and the accuracy scores reported in the experiments. The second is cosine-consistency voting: 5–20 candidate responses from fine-tuned PaliGemma are encoded by the fine-tuned all-MiniLM-L6-v2 (≈80 MB) embedding model, and the candidate whose embedding has the highest average cosine similarity to all other candidates is selected as the final diagnosis.

What would settle it

Take the 90 evaluation images and a pool of human plant pathologists; have them score each greedy and consensus response independently, then compare human agreement with o1-mini scores. If human judges do not prefer the consensus responses at rates comparable to the o1-mini scores (or if o1-mini's labels systematically contradict expert ratings), the central claim of improved diagnostic accuracy fails.

Watch

Extended reading notes

Core claim

The central discovery, stated on the paper's own terms, is that domain-adapted semantic consensus beats single-shot generation for agricultural VLM outputs. When a fine-tuned PaliGemma model samples multiple responses and a fine-tuned all-MiniLM-L6-v2 embedding model selects the response with the highest average pairwise cosine similarity, the chosen response is correct more often than greedy decoding at every evaluated stage. The largest gains appear in the hardest stages—symptom analysis and treatment recommendation—where ordinary text-overlap metrics are shown to be misleading, because the consensus is computed in an embedding space that treats clinically equivalent phrasings as close.

Load-bearing premise

The entire pipeline—both the embedding model's training labels and the final accuracy numbers—depends on the o1-mini prompt-based scorer matching what a real plant pathologist would say is correct; if that scorer is biased, the reported gains are scoring artifacts rather than real diagnostic improvements.

Editorial extensions

If this is right

  • A field team can run the system on a mid-range phone without internet and still get the reported accuracy gains, because voting uses an 80 MB embedding model rather than repeated calls to a large language model.
  • Peak accuracy occurs at 10–15 candidate generations, so deployment can cap the compute per image at that range without losing accuracy.
  • The framework transfers in principle to other crops and other diseases, since nothing in the voting mechanism is specific to maize beyond the fine-tuning data and the scorer prompt.
  • The paper argues that more accurate treatment recommendations enable targeted fungicide application, reducing chemical inputs by 15–30% compared with blanket treatments.

Reading between the lines

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

  • A testable extension the paper leaves implicit: replace the single o1-mini scorer with several different LLM scorers and check whether their consensus better predicts human pathologist agreement; if it does, the evaluation protocol becomes less dependent on one model's biases.
  • The method points to a general recipe for high-stakes, specialist domains where surface text overlap is a poor proxy for correctness: fine-tune a small embedding on domain-expert similarity labels and vote over multiple generations, leaving the generative model untouched.
  • The 0.8 correctness threshold is a free parameter in the paper; a deployment would need to tune it to local treatment costs, since a low threshold risks recommending wrong treatments and a high threshold rejects many valid alternatives.
  • Because only 10 images were used to fine-tune the embedding model, a direct follow-up is to test whether more similarity-labeled images widen the gap between fine-tuned and non-fine-tuned voting or shrink it.
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

4 major / 5 minor

Summary. The paper proposes a domain-aware framework for vision-language model (VLM) based crop disease management. It combines a prompt-based LLM evaluator (o1-mini configured as an expert plant pathologist) with a cosine-consistency self-voting mechanism: multiple candidate responses are generated from a fine-tuned PaliGemma model, encoded with a fine-tuned all-MiniLM-L6-v2 embedding model, and the candidate with highest average pairwise cosine similarity is selected. On a 90-image maize leaf blight test set, the authors report accuracy gains from 82.2% to 87.8% for disease diagnosis, 38.9% to 52.2% for symptom analysis, and 27.8% to 43.3% for treatment recommendation. The claimed contributions are the prompt-based scoring protocol for domain-aware evaluation and a lightweight voting mechanism suitable for mobile deployment.

Significance. If the reported gains were validated against independent expert ground truth, the paper would make a useful practical contribution: it demonstrates that multi-response consensus with a domain-adapted embedding model can improve a small VLM's outputs at low deployment cost. The combination of prompt-based scoring labels with embedding fine-tuning is a reasonable idea, and the three-stage evaluation (diagnosis, symptom analysis, treatment) is clearly structured. However, the significance is currently conditional because every headline number is produced by the same o1-mini prompt-based scorer that generated the ground-truth format and the embedding training labels. The paper provides no human-expert validation of the scorer, no inter-rater agreement, no confidence intervals, and no public data or code, so the central claim cannot be accepted as evidence of real agronomic improvement.

major comments (4)
  1. [Sections 3.4, 4.1, 4.2] The evaluation is circular. Section 4.1 states that o1-mini transformed raw farmer chat interactions into the standardized four-step ground-truth format; Section 3.2 uses the same o1-mini prompt-based scorer to judge generated outputs; Section 3.4 uses those scores as training labels for the embedding model. The test set is then scored with the same rubric (Section 4.2). Because the judge, the label generator, and the training-target generator are all the same model/prompt family, the reported 82.2%->87.8%, 38.9%->52.2%, and 27.8%->43.3% gains measure agreement with o1-mini's stylistic and rubric preferences, not with plant-pathology ground truth. The authors must provide independent validation of the scorer, for example by showing agreement with a panel of human plant pathologists on a held-out subsample and by reporting the scorer's precision/recall against that panel. Without this, the central claim is not established.
  2. [Section 4.1] The dataset description is internally inconsistent with the claim of expert validation. Section 4.1 says each case includes 'expert validation of disease identification and treatment recommendations,' but the only validation step described in the text is the o1-mini transformation of chat interactions into the four-step format. No human pathologist is named, no validation protocol is described, and no inter-rater agreement is reported. If the ground truth is o1-mini-generated, then the accuracy numbers in Table 2 are at best self-consistency scores, not accuracy relative to agronomic reality. The authors need to clarify exactly which parts of the ground truth came from human experts and which from o1-mini, and ideally release a sample of human-labeled cases.
  3. [Table 2 and Section 5.1] The test set is extremely small (90 images) and no uncertainty quantification is provided. The headline treatment gain of 15.5 percentage points corresponds to 14 images (39 vs 25 correct out of 90). With 90 binomial observations, the 95% confidence interval for a 27.8% rate is roughly +/-9 percentage points, and for 43.3% it is roughly +/-10 percentage points, so the reported differences are well within the range of sampling noise. The authors should report confidence intervals or a paired significance test such as McNemar's test for each step and each generation count. The current tables do not support the claim of 'significant progress' in Section 5.1.
  4. [Sections 5.3 and 4.2] The robustness analysis in Section 5.3 only demonstrates that the voting result is stable across generation counts; it does not validate the correctness metric. The 'Winners %' column in Table 2 is consistently lower than the corresponding greedy baseline for Step 2 and Step 3 (e.g., 33.8% winners vs 38.9% greedy at 20 generations), yet the selected FT response is reported as higher. This is not necessarily wrong, but the paper does not analyze the selection mechanism (e.g., how often the chosen candidate is the correct one, or how embedding fine-tuning changes candidate ranking). The authors should add an analysis of selection accuracy and a comparison of the selected response versus the best candidate; otherwise the gain could be an artifact of evaluating only the selected response.
minor comments (5)
  1. [Section 3.5] The formatting of the numbered multi-response generation steps is broken: '1. Multi-Response Generation... 2. Domain-Aware Embedding... 3. Consensus Calculation' is run together in one paragraph, making it difficult to see the pipeline.
  2. [Table 2] The values for 'Winners %' are not explained precisely enough (e.g., whether the denominator is the total number of generated candidates across all test images). A definition in the caption or in Section 4.2 would help.
  3. [Section 5.3] The text says 'Peak performance typically occurs with 10-15 generations,' but Table 2 shows the best FT result for Step 1 at 5 generations (87.8%), Step 2 at 15 generations (52.2%), and Step 3 at 20 generations (43.3%). A single summary statement should be supported by the table or qualified.
  4. [References] Reference [15] cites the o1 preview announcement from 2023, but the paper uses o1-mini; a citation that directly describes o1-mini and the evaluation date would be more precise.
  5. [General] The paper does not state whether data, code, or prompts will be released; the single supplementary URL for the prompt is not sufficient for reproducibility of the evaluation pipeline.

Circularity Check

2 steps flagged · score 6.0 of 10

The headline accuracy gains are measured by the same o1-mini prompt-based scorer that generated the training labels for the voting embedding, so the evaluation reduces partly to self-agreement with that scorer.

  1. fitted input called prediction [Section 3.4 (Embedding Model Fine-tuning), Section 3.2 (Prompt-Based Scoring), Section 4.2 (Evaluation Methodology)]
    "This framework is never invoked at inference time. Instead, it provides high-quality semantic-similarity targets for fine-tuning the embedding model and reliable, domain-aware evaluation scores. ... Target similarity labels are assigned based on prompt-based scoring: Both outputs ≥ 0.8 ⇒ label = 1.0 ... Responses are considered correct when achieving prompt-based scoring ≥ 0.8, indicating equivalence to expert-level agricultural advice."

    The same o1-mini prompt-based scorer is used twice: it creates the similarity targets that fine-tune the voting embedding (Section 3.4), and it defines correctness in the evaluation (Section 4.2). The cosine-consistency selector is therefore optimized to reproduce o1-mini's score function, and the reported accuracies then measure how often the selected response satisfies that same function. The gains (82.2% to 87.8%, 38.9% to 52.2%, 27.8% to 43.3%) are consequently a measure of alignment between the trained embedding and o1-mini's rubric, not an independently validated agronomic improvement. No human-expert validation, inter-rater agreement, or error bars are reported for the ≥ 0.8 threshold.

  2. other [Section 4.1 (Agricultural Dataset) and Section 4.2 (Evaluation Methodology)]
    "We utilized o1-mini to transform original chat-based interactions into a standardized four-step format (Assessment, Analysis, Treatment, and Prevention), each separated by semicolons. ... Responses are considered correct when achieving prompt-based scoring ≥ 0.8, indicating equivalence to expert-level agricultural advice."

    The ground-truth reference format and the correctness judge are both produced by o1-mini. Because the reference is itself a model-generated standardization of chat interactions, both sides of the comparison share o1-mini's formatting and rubric preferences. The reported correctness therefore measures agreement with o1-mini's reformatting and scoring behavior, rather than agreement with an independent expert-created gold standard, making the evaluation a self-agreement loop between two calls to the same model.

full rationale

The central accuracy claim is not mathematically forced: PaliGemma must still generate biologically plausible candidates, and voting can fail even when the scorer is used as the metric. However, the evaluation is not independent. The o1-mini prompt-based scorer supplies the regression targets for the embedding model that drives FT Voting, and the same scorer then declares responses correct at the ≥ 0.8 threshold. This is a feedback loop: the selector is trained to approximate the scorer and then scored by it. The robustness analysis (Section 5.3) only varies the number of generations and does not validate the metric against external expert labels. No code, dataset, or human-annotation comparison is provided, so the reviewer cannot separately confirm that the gains reflect agronomic correctness rather than stylistic alignment with o1-mini. Because the consensus mechanism itself has independent behavioral content, the circularity is partial, not total, and the paper is not a pure restatement of its inputs.

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

The framework introduces no new physical entities. Its load-bearing assumptions are all domain assumptions about the validity of the LLM-based scorer and the representativeness of the proprietary dataset.

free parameters (4)
  • Correctness threshold = 0.80
    Section 4.2 counts responses with prompt-based score >= 0.8 as correct. This hand-chosen threshold directly determines all reported accuracy percentages.
  • Embedding training label thresholds = 1.0 / 0.8 / 0.1
    Section 3.4 assigns target labels based on whether both, one, or neither paired output scores >= 0.8. These cutoffs are arbitrary and shape the embedding model.
  • Sampling temperature = 1.0
    Section 3.5 uses temperature 1.0 for multi-response generation. The paper does not explore other temperatures.
  • Generation count = 5, 10, 15, 20 (peak at 10-15)
    Section 5.3 states peak performance at 10-15 generations; the count is a tuning knob rather than a single fitted constant.
assumptions (4)
  • domain assumption o1-mini prompt-based scoring is a valid proxy for expert plant pathologist assessment
    Section 3.2 introduces the prompt-based score as ground truth for correctness, and Section 4.2 uses it to label responses correct. No human-expert validation is provided.
  • domain assumption The selected 500 Kissan.ai interactions and the 90-sample evaluation set represent field conditions
    Section 4.1 describes the dataset as farmer-captured, but gives no details on how the 500 of 826 interactions were selected or whether the 90 eval images are diverse. Selection bias is possible.
  • domain assumption Fine-tuning the embedding on 2100 LLM-labeled pairs improves semantic similarity in the agricultural domain
    Section 3.4 relies on this assumption; no independent measure of embedding quality is given.
  • domain assumption Cosine similarity in the fine-tuned embedding space is a reliable signal for consensus
    Section 3.5 assumes the most central response is the most correct; the 'Winners %' column partially supports this but it is not tested statistically.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Self-Consistency in Vision-Language Models for Precision Agriculture: Multi-Response Consensus for Crop Disease Management." pith.science (2026). https://pith.science/paper/X5432EQT

@misc{pith2026250708024,
  author       = {Pith},
  title        = {Pith review of: Self-Consistency in Vision-Language Models for Precision Agriculture: Multi-Response Consensus for Crop Disease Management},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/X5432EQT}},
  note         = {Machine review of arXiv:2507.08024}
}
read the original abstract

Precision agriculture relies heavily on accurate image analysis for crop disease identification and treatment recommendation, yet existing vision-language models (VLMs) often underperform in specialized agricultural domains. This work presents a domain-aware framework for agricultural image processing that combines prompt-based expert evaluation with self-consistency mechanisms to enhance VLM reliability in precision agriculture applications. We introduce two key innovations: (1) a prompt-based evaluation protocol that configures a language model as an expert plant pathologist for scalable assessment of image analysis outputs, and (2) a cosine-consistency self-voting mechanism that generates multiple candidate responses from agricultural images and selects the most semantically coherent diagnosis using domain-adapted embeddings. Applied to maize leaf disease identification from field images using a fine-tuned PaliGemma model, our approach improves diagnostic accuracy from 82.2\% to 87.8\%, symptom analysis from 38.9\% to 52.2\%, and treatment recommendation from 27.8\% to 43.3\% compared to standard greedy decoding. The system remains compact enough for deployment on mobile devices, supporting real-time agricultural decision-making in resource-constrained environments. These results demonstrate significant potential for AI-driven precision agriculture tools that can operate reliably in diverse field conditions.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 27 canonical work pages

  1. [1]

    Central to this trans- formation is the automated analysis of agricultural images for real-time crop monitoring, disease detection, and treatment recommendation

    INTRODUCTION Precision agriculture represents a paradigm shift toward data- driven farming practices that optimize resource utilization while maintaining crop productivity [1]. Central to this trans- formation is the automated analysis of agricultural images for real-time crop monitoring, disease detection, and treatment recommendation. Climate change has...

    work page

  2. [2]

    Prompt-based agricultural expert evaluation: A scalable assessment protocol that configures a language model as an expert plant pathologist, enabling domain-aware evaluation of image analysis outputs without requiring human experts

    work page

  3. [3]

    Multi-response self-consistency for agricultural im- ages: A cosine similarity-based voting mechanism that pro- cesses multiple candidate interpretations of agricultural im- ages and selects the most reliable diagnosis through semantic consensus. Our framework demonstrates substantial improvements across all aspects of maize leaf disease identification us...

    work page Pith review arXiv 2025

  4. [4]

    RELATED WORK Agricultural image analysis has progressed from hand-crafted colour/texture descriptors to deep CNNs [4], and most re- cently to vision–language models (VLMs) that attach natural- language symptom descriptions and treatment advice to crop images [5]. Early domain-specific efforts such as WDLM, built on CLIP for wheat rust recognition [6], and...

    work page

  5. [5]

    maize leaf blight, severe infection symp- toms: tan to grayish spots with darker borders, analysis: se- vere tan spots require mancozeb

    METHODOLOGY 3.1. Domain-Specific Evaluation Challenges Evaluating generative model outputs for agricultural disease management requires domain-specific expertise that tradi- tional metrics fail to capture. We examined three traditional metrics and identified critical limitations: Embedding-Based Metrics: Off-the-shelf cosine simi- larity focuses on vector...

    work page

  6. [6]

    Extract key characteristics for each description: disease identification, progression, symptoms, and treatment im- plications

    work page

  7. [7]

    Compare characteristics point-by-point, noting similari- ties and differences

    work page

  8. [8]

    spots” vs. “le- sions

    Handle synonymous terminology (e.g., “spots” vs. “le- sions”) so wording differences do not mask clinical equivalence

    work page

Show all 31 references
  1. [9]

    Assess overall disease impact and severity, determining whether fundamental disease presentation and required treatment are equivalent

  2. [10]

    Return a dictionary with score and reasoning. The similarity score follows this rubric: • 0.00–0.50: Different diseases or requiring different treat- ments 1Full prompt text available at https://sigport.org/ documents/supplmental-o1-mini-prompt . • 0.51–0.79: Same disease but ...

  3. [11]

    Domain-Aware Embedding: Encode each response using the agricultural domain-adapted embed- ding model 3

    Multi-Response Generation: Generate diverse can- didate analyses using temperature sampling ( t = 1.0) from the fine-tuned PaliGemma model to explore different diag- nostic possibilities 2. Domain-Aware Embedding: Encode each response using the agricultural domain-adapted embe...

  4. [12]

    Winners %

    EXPERIMENTAL SETUP 4.1. Agricultural Dataset The dataset was sourced from Kissan.AI’s mobile application chat interactions with farmers. Farmer-uploaded plant leaf images were structured around three key aspects: disease de- tection with symptom analysis, fungicide recommendat...

  5. [13]

    RESULTS AND ANALYSIS Table 2 presents comprehensive results across agricultural image processing stages, demonstrating consistent improve- ments through domain-aware self-consistency applied to fine- tuned PaliGemma outputs. 5.1. Agricultural Domain Performance The domain-adap...

  6. [14]

    DISCUSSION AND CONCLUSIONS Our domain-aware framework achieves measurable improve- ments in agricultural image analysis across all evaluated stages. These improvements directly enable precision farm- ing practices: accurate disease diagnosis supports targeted fungicide applica...

  7. [15]

    Precision agricul- ture: A worldwide overview,

    N. Zhang, M. Wang, and N. Wang, “Precision agricul- ture: A worldwide overview,” Computers and electron- ics in agriculture, vol. 36, no. 2-3, pp. 113–132, 2002

    2002

  8. [16]

    Under- standing small-scale farmers’ perception and adaption strategies to climate change impacts: Evidence from two agro-ecological zones bordering national parks of uganda,

    C. E. Tiyo, F. Orach-Meza, and E. L. Edroma, “Under- standing small-scale farmers’ perception and adaption strategies to climate change impacts: Evidence from two agro-ecological zones bordering national parks of uganda,” Journal of Agricultural Science, vol. 7, no. 10, p. 253, 2015

    2015

  9. [17]

    Pairwise feature learning for unseen plant disease recognition,

    A. Y . Hao Chai, S. Han Lee, F. S. Tay, Y . Lung Then, H. Go ¨eau, P. Bonnet, and A. Joly, “Pairwise feature learning for unseen plant disease recognition,” in 2023 IEEE International Conference on Image Processing (ICIP), pp. 306–310, 2023

    2023

  10. [18]

    Using deep learning for image-based plant disease detection,

    S. P. Mohanty, D. P. Hughes, and M. Salath ´e, “Using deep learning for image-based plant disease detection,” Frontiers in plant science, vol. 7, p. 1419, 2016

    2016

  11. [19]

    Learning transferable visual models from nat- ural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, et al. , “Learning transferable visual models from nat- ural language supervision,” in International conference on machine learning, pp. 8748–8763, PMLR, 2021

    2021

  12. [20]

    Visual large language model for wheat disease diagno- sis in the wild,

    K. Zhang, L. Ma, B. Cui, X. Li, B. Zhang, and N. Xie, “Visual large language model for wheat disease diagno- sis in the wild,” Computers and Electronics in Agricul- ture, vol. 227, p. 109587, 2024

    2024

  13. [21]

    A framework for agricultural intelli- gent analysis based on a visual language large model,

    P. Yu and B. Lin, “A framework for agricultural intelli- gent analysis based on a visual language large model,” Applied Sciences, vol. 14, no. 18, p. 8350, 2024

    2024

  14. [22]

    Self- consistency improves chain of thought reasoning in lan- guage models,

    X. Wang, J. Wei, D. Schuurmans, Q. Le, E. Chi, S. Narang, A. Chowdhery, and D. Zhou, “Self- consistency improves chain of thought reasoning in lan- guage models,” arXiv preprint arXiv:2203.11171, 2022

    2022 arXiv

  15. [23]

    Chain-of-thought prompting elicits reasoning in large language models,

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V . Le, D. Zhou, et al. , “Chain-of-thought prompting elicits reasoning in large language models,” Advances in neural information processing systems , vol. 35, pp. 24824–24837, 2022

    2022

  16. [24]

    Uni- versal self-consistency for large language model gener- ation,

    X. Chen, R. Aksitov, U. Alon, J. Ren, K. Xiao, P. Yin, S. Prakash, C. Sutton, X. Wang, and D. Zhou, “Uni- versal self-consistency for large language model gener- ation,” arXiv preprint arXiv:2311.17311, 2023

    2023 arXiv

  17. [25]

    AgroGPT: Efficient agricultural vision–language model with expert tuning,

    M. Awais, A. H. S. A. Alharthi, A. Kumar, H. Cholakkal, and R. M. Anwer, “AgroGPT: Efficient agricultural vision–language model with expert tuning,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , IEEE, 2025. arXiv:2410.08405

    2025 arXiv

  18. [26]

    Sentence-bert: Sentence embeddings using siamese bert-networks,

    N. Reimers and I. Gurevych, “Sentence-bert: Sentence embeddings using siamese bert-networks,” in Proceed- ings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th Interna- tional Joint Conference on Natural Language Process- ing (EMNLP-IJCNLP...

    2019

  19. [27]

    Bleurt: Learning robust metrics for text generation,

    T. Sellam, D. Das, and A. P. Parikh, “Bleurt: Learning robust metrics for text generation,” 2020

    2020

  20. [28]

    Evaluating deep learning techniques for natural lan- guage inference,

    P. Eleftheriadis, I. Perikos, and I. Hatzilygeroudis, “Evaluating deep learning techniques for natural lan- guage inference,” Applied Sciences, 2023

    2023

  21. [29]

    Introducing openai o1 pre- view

    OpenAI, “Introducing openai o1 pre- view.” https://openai.com/index/ introducing-openai-o1-preview/ , 2023. Accessed: 2024-12-28

    2023

  22. [30]

    Paligemma: A versatile 3b vlm for transfer,

    L. Beyer, A. Steiner, A. S. Pinto, A. Kolesnikov, X. Wang, D. Salz, M. Neumann, I. Alabdulmohsin, M. Tschannen, E. Bugliarello, et al. , “Paligemma: A versatile 3b vlm for transfer,” arXiv preprint arXiv:2407.07726, 2024

    2024 arXiv

  23. [31]

    all-MiniLM-L6-v2: A lightweight sentence transformer model

    F. Wang, Y . Cheng, and S. Gan, “all-MiniLM-L6-v2: A lightweight sentence transformer model.” https:// huggingface.co/sentence-transformers/ all-MiniLM-L6-v2 , 2021

    2021

Pith tools

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