First Train to Generate, then Generate to Train: UnitedSynT5 for Few-Shot NLI

Anush Mahajan; Ayushi Agarwal; Eishkaran Singh; Sourav Banerjee

REVIEW 5 major objections 4 minor 1 cited by

First Train to Generate, then Generate to Train: UnitedSynT5 for Few-Shot NLI

T0 review · 5 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper claims that augmenting NLI training with T5-generated, teacher-filtered premise-hypothesis pairs reaches 94.7% accuracy on SNLI, 94.0% on E-SNLI, and 92.6% on MultiNLI, surpassing previous state-of-the-art results.

desk verdict The reported SOTA numbers are new, but the paper's central claim that the synthetic pipeline causes the gains is unsupported by the experimental design. read the letter →

arxiv 2412.09263 v2 pith:QXJ3JJXH submitted 2024-12-12 cs.CL cs.AI

Sourav Banerjee , Anush Mahajan , Ayushi Agarwal , Eishkaran Singh This is my paper · ORCID

classification cs.CLcs.AI

keywords NaturalLanguageInferenceSyntheticDataAugmentationFew-ShotLearningEntailment(EFL)T5GTR-T5-XLSNLIMultiNLI

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 sets out to show that the accuracy ceiling in natural language inference is not set by model architecture alone but by the limited variety of the human-labeled training data. It augments the SNLI training set with machine-generated premise-hypothesis pairs: a FLAN-T5 XL generator writes new hypotheses, a GTR-T5-XL teacher keeps only examples whose label it agrees with, and the surviving 467,683 pairs are reformatted in the EFL style with the label stated inside the hypothesis. A GTR-T5-XL model trained on the combined data reaches 94.7% accuracy on SNLI, 94.0% on E-SNLI, and 92.6% on MultiNLI, each above the previous best numbers the paper lists. If the result holds, it would mean that synthetic, teacher-filtered data can substitute for part of the human annotation effort in NLI and in related language-understanding tasks.

What carries the argument

The load-bearing mechanism is the generation-filter-retrain loop. A FLAN-T5 XL generator (3B parameters) produces hypotheses conditioned on a premise and two few-shot examples; a label-alignment filter keeps only pairs where the GTR-T5-XL teacher's prediction equals the generation label; and the EFL conversion, which phrases the label inside the hypothesis, turns the three-way classification into a binary verification task. The size of the retained synthetic set is what expands the training distribution beyond the original 570,000 human pairs.

What would settle it

Train the same pipeline with the filter inverted, keeping only synthetic examples whose teacher prediction disagrees with the generator's label, and compare SNLI test accuracy; if accuracy does not drop, the label-alignment filter is not the source of the reported gain. A direct check is to have human annotators label a random sample of the 467,683 retained pairs and measure how often the generated label matches human judgment.

Watch

Extended reading notes

Core claim

The central claim is that a generation-filter-retrain loop improves few-shot NLI beyond the prior state of the art. Starting from 95% of SNLI (521,898 premises) and two randomly sampled few-shot examples per prompt, the FLAN-T5 XL generator produces a candidate hypothesis for each premise; the cleaning step discards 54,216 candidates whose label does not match the GTR-T5-XL teacher or that duplicate existing training examples, leaving 467,683 synthetic pairs. These are converted to EFL form so that each hypothesis reads as "The hypothesis ... is a [label] of the premise," and they are mixed with the human-labeled training set to fine-tune a GTR-T5-XL model. The reported outcome is 94.7% test accuracy on SNLI, 94.0% on E-SNLI, and 92.6% on MultiNLI, surpassing the 93.1% EFL baseline on SNLI and the other comparison numbers in the paper's tables.

Load-bearing premise

The load-bearing premise is that a synthetic example improves training whenever the teacher model's predicted label matches the generator's label; if the teacher has systematic blind spots, the filter will amplify those blind spots rather than introduce diverse, independent examples.

Editorial extensions

If this is right

If the reported numbers hold, NLI accuracy is no longer bounded by the architecture; augmenting the training distribution with generated, teacher-filtered examples raises test accuracy above the prior state of the art.
The same pipeline transfers to E-SNLI and MultiNLI, with 94.0% and 92.6% accuracy, indicating the benefit is not specific to SNLI's single genre.
Scaling the classifier matters: the 335M-parameter GTR-Large version reaches 93.5% on SNLI while the 3B-parameter GTR-T5-XL reaches 94.7%, so larger models absorb more of the augmented data's benefit.
Because the synthetic data is converted into EFL format, existing EFL training pipelines can consume generated examples without changing their learning objective.

Reading between the lines

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

The paper does not ablate the label-alignment filter, so the reported gain could come partly from distilling the teacher's own preferences rather than from genuinely new linguistic diversity; a comparison against a teacher-disagreement filter would separate those effects.
The 95/5 split and the choice of two few-shot examples per prompt are untested design choices; varying them is a natural next step and may change both the cost and the accuracy.
The paper's own Section 5 notes that generation cost grows with prompt length and that smaller generators lose coherence on long prompts, so the two-example prompt is a practical constraint as well as a design choice.
The same generation-filter-retrain recipe could be applied to other label-conditioned tasks such as fact verification or reading comprehension whenever a teacher model can supply agreement labels.

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.

Desk Editor's Note

read the letter

Dear Colleague,

The headline numbers are new, but the paper's central claim—that the synthetic data pipeline causes the gains—is not supported by the experiments. The recipe is clearly described, and the authors get credit for that. They take EFL, add a T5 generator to produce new premise-hypothesis pairs, filter by label agreement with a GTR-T5-XL teacher and by novelty, then train a GTR-T5-XL classifier. The pseudo-code and worked examples make the pipeline concrete.

The soft spots are load-bearing. There is no same-backbone baseline. Every UnitedSynT5 run uses a GTR backbone, and the only non-augmented comparison is EFL with a different architecture (T5-XXL). The fact that the 335M model already reaches 93.5%, beating EFL's 93.1% with a much larger model, suggests the backbone choice could explain most of the gain. The paper attributes results to "model scaling and synthetic data augmentation" but never separates the two. Without a GTR-T5-XL fine-tuned on the original data alone, there is no evidence the augmented examples do anything.

Second, the label-alignment filter is circular. The teacher (GTR-T5-XL) decides which synthetic examples are kept, and that same model is then fine-tuned on the kept set. That is self-confirmation, not new information. If the teacher has systematic biases, the filter amplifies them. The limitation section doesn't address this.

Third, reporting is thin: no error bars, no ablations of the filter or generator, no code or data. The baseline comparisons are not apples-to-apples—different models, different sizes, different settings. There's also an internal inconsistency: Algorithm 1 uses a label L for generated examples but never defines where it comes from, and the text claims each example yields three EFL instances while the algorithm produces only one. That L is load-bearing, because the filter's correctness depends on it.

The paper's own limitations section mentions computational cost and prompt length but not the attribution problem. So the central promise is untested. As a SOTA claim, I wouldn't trust it. As a starting point for synthetic augmentation in NLI, the recipe is worth redoing with proper controls.

I'd send this to a referee, but with a request for major revisions: add a same-backbone no-synthetic baseline, report error bars, fix the undefined label, and discuss the filter's circularity. I would not cite it in its current form.

Recommendation: engage with it as a work-in-progress, not as a reliable result.

Referee Report

5 major / 4 minor

Summary. The paper proposes UnitedSynT5, a synthetic data augmentation method for few-shot Natural Language Inference (NLI). The method uses a FLAN-T5 XL generator to produce additional premise-hypothesis pairs, filters these for label alignment with a GTR-T5-XL teacher and for redundancy, converts the retained examples into the EFL (Entailment as Few-Shot Learner) format, and fine-tunes a GTR-T5-XL classifier on the augmented data. The authors report new state-of-the-art accuracies on SNLI (94.7%), E-SNLI (94.0%), and MultiNLI (92.6%), claiming that the augmentation pipeline improves over prior methods such as EFL.

Significance. If the reported results are reproducible and robust, the paper would demonstrate a practical way to push NLI accuracy beyond the current human-annotated data ceiling, which is a significant contribution. The combination of a T5-based generator, teacher-filtered synthetic data, and the EFL reformulation is a reasonable design, and the paper includes a step-by-step algorithm and results on three benchmarks. However, the significance is contingent on resolving the experimental and methodological issues below; as written, the evidence does not establish that the synthetic augmentation, rather than the choice of backbone or the filtering procedure, causes the reported gains.

major comments (5)

[Section 4 and Table 2] No same-backbone baseline is provided. All UnitedSynT5 results use GTR-Large or GTR-T5-XL classifiers, while the only non-augmented comparison, EFL, uses T5-XXL. There is no experiment fine-tuning GTR-Large or GTR-T5-XL on the original SNLI, E-SNLI, or MultiNLI training data without synthetic examples; thus the reported gains over EFL (e.g., 93.5% vs. 93.1% for the 335M model) cannot be attributed to the synthetic augmentation pipeline rather than to the choice of backbone and training setup. Section 4's claim that the results 'highlight the effectiveness of model scaling and synthetic data augmentation' conflates these two factors, and without a factorial design the paper's central claim that augmentation improves accuracy is unsupported.
[Section 3.3.1 and Algorithm 1 Step 2] The label-alignment filter retains only synthetic examples for which the teacher model GTR-T5-XL's prediction matches the generation label. Because the same model family is then fine-tuned on this filtered data, the filter reinforces the teacher's existing biases and the synthetic examples do not provide independent evidence for new generalizations. The paper does not report ablations with unfiltered synthetic data, with a different teacher, or with a held-out teacher, so the claimed benefit of this cleaning step is not demonstrated. This is a load-bearing methodological issue for the claim that the pipeline adds diverse, high-quality training data.
[Section 3.1 and Algorithm 1] The label variable L used in Step 2 is never defined in Step 1. Step 1 stores only (P, H) for each generated hypothesis, but Step 2 iterates over (P, H, L) and compares L to the teacher prediction. While Section 3.2.1's prompt includes a label (e.g., 'Your Task: Label: Contradiction'), the algorithm and the text's claim that generation proceeds 'without requiring manual labeling' are inconsistent with the use of generator-provided labels in filtering. This ambiguity makes the data generation and cleaning process irreproducible as written.
[Table 2] The paper reports single accuracy numbers without error bars, number of seeds, or significance tests. The claimed improvements over previous SOTA are small (e.g., 93.5% vs. 93.1% for the 335M model on SNLI), and without variance estimates it is impossible to tell whether these differences are meaningful. Especially for 3B-scale fine-tuning, run-to-run variation could easily exceed 0.4 points. The absence of any statistical reliability measure undermines the SOTA claims.
[Sections 3.5.2 and 4] The evaluation protocol is underspecified. Section 3.5.2 states only that the model is 'evaluated on the original SNLI dataset' without specifying the split (train/validation/test) or the number of runs. Moreover, the paper reports results on E-SNLI and MultiNLI, but the methodology describes synthetic data generation only for SNLI; it is unclear whether separate synthetic corpora were generated for each benchmark, whether the SNLI synthetic data were reused, and what few-shot configuration was used for E-SNLI and MultiNLI. Without this information, the cross-benchmark generalization claims cannot be assessed.

minor comments (4)

[Section 3.4] The text claims 'Each original example yields three EFL-formatted instances, one for each possible label' and states |DEFL| = 3 × |Doriginal|, but Algorithm 2 constructs exactly one HEFL per (P,H,L). Clarify whether the training data actually includes three variants per premise or only the gold-label variant.
[Sections 3.2 and 3.2.2] The generator is described both as being 'trained' with cross-entropy loss and as using a few-shot learning approach with two examples in the prompt. Since FLAN-T5 XL is instruction-tuned, the paper should state whether any generator parameters were updated or whether it was used purely in a prompted, zero-gradient mode.
[Table 2] The E-SNLI row for UnitedSynT5 (335M) reports 89.8%, which is below BART-large's 92.3%; the text's statement that the approach 'broke previous records' on E-SNLI applies only to the 3B model and should be phrased to avoid overclaiming.
[Abstract and Section 3.4] Typos and notation: the abstract has 'addressthis' (missing space), Table 2 has a stray 'r' at the end, and the EFL string in Section 3.4 reads 'is a + L + of the premise' with misplaced plus signs; these should be corrected.

Circularity Check

0 steps flagged · score 0.0 of 10

No definitional circularity; the teacher-filtered synthetic data is self-training but not a circular derivation, and the missing same-backbone baseline is an experimental-design issue.

full rationale

The paper's derivation chain is: generate synthetic premise-hypothesis-label triples with FLAN-T5 XL, filter them by requiring agreement with a GTR-T5-XL teacher (Algorithm 1 Step 2), convert to the EFL format, train a new GTR-T5-XL model on the selected set plus original SNLI data, and evaluate on the held-out test set. No equation in the paper defines its prediction in terms of its training input, and no fitted parameter is renamed as a prediction. The label-alignment filter does mean that the retained synthetic labels are exactly those the teacher model already endorses, so the augmentation cannot introduce labels that contradict the teacher; this is a real self-training and confirmation concern that may limit diversity or amplify teacher biases. However, it is not a circularity in the required sense: the student model is still trained on original human labels and evaluated against held-out human labels, and the reported test accuracy is not an algebraic or statistical consequence of the filtering criterion. The absence of a same-backbone, no-synthetic baseline means the paper cannot cleanly attribute its gains to augmentation rather than to GTR-T5-XL scale and training setup, but that is an attribution and experimental-design limitation, not a circular derivation. The only self-citations ([15] for few-shot learning background and [23] for synthetic-data generation context) are not load-bearing; the proposed method is described in the paper itself. Thus no circular step meeting the quoted-evidence bar can be identified.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central contribution rests on the assumption that generated data filtered by a teacher's own predictions is informative. This is a fragile premise because it can introduce selection bias. No independent evidence, such as a distinct held-out set used to validate the synthetic data, is provided.

free parameters (4)

95%-5% generation/few-shot split = 95:5
Chosen without ablation; determines how many premises are used for generation and how many support examples are provided in the prompt.
Number of few-shot examples in prompt = 2
Chosen without ablation; affects generation quality and prompt length.
Classifier architecture = 3 FC layers, GeLU, dropout 0.1
These settings are stated without justification or comparison.
Backbone model size = 335M vs 3B
The choice of GTR-Large vs GTR-T5-XL affects the reported accuracy, and the reason for the two variants is not discussed.

assumptions (3)

domain assumption Synthetic examples filtered by teacher-model agreement improve NLI generalization.
The paper assumes this without an ablation or theoretical justification; it is the central premise of the augmentation pipeline.
domain assumption FLAN-T5 XL can generate valid premise-hypothesis pairs for NLI from the provided prompts.
The paper does not validate the quality of generated hypotheses beyond the teacher filter.
ad hoc to paper The GTR-T5-XL architecture can be repurposed as an NLI classifier with a three-layer head.
This design is introduced for this paper and is not a standard use of GTR-T5, which is a dual-encoder retriever.

how reviews work

0 comments

Cite this review

Pith. "Pith review of First Train to Generate, then Generate to Train: UnitedSynT5 for Few-Shot NLI." pith.science (2026). https://pith.science/paper/QXJ3JJXH

@misc{pith2026241209263,
  author       = {Pith},
  title        = {Pith review of: First Train to Generate, then Generate to Train: UnitedSynT5 for Few-Shot NLI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QXJ3JJXH}},
  note         = {Machine review of arXiv:2412.09263}
}

read the original abstract

Natural Language Inference (NLI) tasks require identifying the relationship between sentence pairs, typically classified as entailment, contradiction, or neutrality. While the current state-of-the-art (SOTA) model, Entailment Few-Shot Learning (EFL), achieves a 93.1% accuracy on the Stanford Natural Language Inference (SNLI) dataset, further advancements are constrained by the dataset's limitations. To address this, we propose a novel approach leveraging synthetic data augmentation to enhance dataset diversity and complexity. We present UnitedSynT5, an advanced extension of EFL that leverages a T5-based generator to synthesize additional premise-hypothesis pairs, which are rigorously cleaned and integrated into the training data. These augmented examples are processed within the EFL framework, embedding labels directly into hypotheses for consistency. We train a GTR-T5-XL model on this expanded dataset, achieving a new benchmark of 94.7% accuracy on the SNLI dataset, 94.0% accuracy on the E-SNLI dataset, and 92.6% accuracy on the MultiNLI dataset, surpassing the previous SOTA models. This research demonstrates the potential of synthetic data augmentation in improving NLI models, offering a path forward for further advancements in natural language understanding tasks.

Figures

Figures reproduced from arXiv: 2412.09263 by the authors.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

Survey of NLU Benchmarks Diagnosing Linguistic Phenomena: Why not Standardize Diagnostics Benchmarks?
cs.CL 2025-07 conditional novelty 4.0 of 10

A survey of NLU diagnostics benchmarks finds no shared naming convention or standard set of linguistic phenomena, and asks whether the field should build an ISO-like evaluation standard.

Reference graph

Works this paper leans on

34 extracted references · 30 canonical work pages · cited by 1 Pith paper

[1]

A large annotated corpus for learning natural language inference,

S. R. Bowman, G. Angeli, C. Potts, and C. D. Manning, “A large annotated corpus for learning natural language inference,” inProceedings of the 2015 Conference on Empirical Methods in Natural Language Processing(L. Màrquez, C. Callison-Burch, and J. Su, eds.), (Lisbon, Portugal), pp. 632–642, Association for Computational Linguistics, Sept. 2015

work page 2015
[2]

A decomposable attention model for natural language inference,

A. Parikh, O. Täckström, D. Das, and J. Uszkoreit, “A decomposable attention model for natural language inference,” inProceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pp. 2249–2255, 2016

work page 2016
[3]

Enhanced lstm for natural language inference,

Q. Chen, X. Zhu, Z. Ling, S. Wei, H. Jiang, and D. Inkpen, “Enhanced lstm for natural language inference,” inProceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 1657–1668, 2017

work page 2017
[4]

Improving language understanding by generative pre-training,

A. Radford, K. Narasimhan, T. Salimans, and I. Sutskever, “Improving language understanding by generative pre-training,” 2018. OpenAI preprint

work page 2018
[5]

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,” 2019

2019
[6]

Xlnet: General- ized autoregressive pretraining for language understanding,

Z. Yang, Z. Dai, Y. Yang, J. Carbonell, R. R. Salakhutdinov, and Q. V. Le, “Xlnet: General- ized autoregressive pretraining for language understanding,”Advances in neural information processing systems, vol. 32, 2019

work page 2019
[7]

Roberta: A robustly optimized bert pretraining approach,

Y. Liu, M. Ott, N. Goyal, J. Du, M. Joshi, D. Chen, O. Levy, M. Lewis, L. Zettlemoyer, and V. Stoyanov, “Roberta: A robustly optimized bert pretraining approach,” inarXiv preprint arXiv:1907.11692, 2019

arXiv 1907
[8]

Albert: A lite bert for self-supervised learning of language representations,

Z. Lan, M. Chen, S. Goodman, K. Gimpel, P. Sharma, and R. Soricut, “Albert: A lite bert for self-supervised learning of language representations,” 2020

work page 2020

Show all 34 references

[9]

Deberta: Decoding-enhanced bert with disentangled attention,

P. He, J. Gao, W.-t. Yih, and X. Deng, “Deberta: Decoding-enhanced bert with disentangled attention,” arXiv preprint arXiv:2106.03654, 2021

2021 arXiv
[10]

Entailment as few-shot learner,

W. Yin, J. Hay, T. Khot, A. Sabharwal, P. Clark, and D. Roth, “Entailment as few-shot learner,” in Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp. 9368–9383, 2021

2021
[11]

Papers with Code - SNLI Benchmark (Natural Language Inference) — paperswithcode.com

“Papers with Code - SNLI Benchmark (Natural Language Inference) — paperswithcode.com.” https://paperswithcode.com/sota/natural-language-inference-on-snli . [Accessed 18-10- 2024]

2024
[12]

An extended model of natural logic,

B. MacCartney and C. D. Manning, “An extended model of natural logic,” inProceedings of the Eight International Conference on Computational Semantics(H. Bunt, ed.), (Tilburg, The Netherlands), pp. 140–156, Association for Computational Linguistics, Jan. 2009

2009
[13]

Natural language inference,

B. MacCartney, “Natural language inference,” inStanford University, 2008

2008
[14]

Deep contextualized word representations,

M. E. Peters, M. Neumann, M. Iyyer, M. Gardner, C. Clark, K. Lee, and L. Zettlemoyer, “Deep contextualized word representations,” inProceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, V...

2018
[15]

Llms will always hallucinate, and we need to live with this,

S. Banerjee, A. Agarwal, and S. Singla, “Llms will always hallucinate, and we need to live with this,” 2024. 12 A preprint - December 16, 2024

2024
[16]

A broad-coverage challenge corpus for sentence understanding through inference,

A. Williams, N. Nangia, and S. Bowman, “A broad-coverage challenge corpus for sentence understanding through inference,” inProceedings of the 2018 Conference of the North American ChapteroftheAssociationforComputationalLinguistics: HumanLanguageTechnologies,Volume 1 (Long Pape...

2018
[17]

Exploring the limits of transfer learning with a unified text-to-text transformer,

C. Raffel, N. Shazeer, A. Roberts, K. Lee, S. Narang, M. Matena, Y. Zhou, W. Li, and P. J. Liu, “Exploring the limits of transfer learning with a unified text-to-text transformer,” 2023

2023
[18]

Large dual encoders are generalizable retrievers,

J. Ni, C. Qu, J. Lu, Z. Dai, G. H. Ábrego, J. Ma, V. Y. Zhao, Y. Luan, K. B. Hall, M.-W. Chang, and Y. Yang, “Large dual encoders are generalizable retrievers,” 2021

2021
[19]

The pascal recognising textual entailment challenge,

I. Dagan, O. Glickman, and B. Magnini, “The pascal recognising textual entailment challenge,” in Machine Learning Challenges Workshop, pp. 177–190, Springer, 2005

2005
[20]

Knowledgeable reader: Enhancing cloze-style reading comprehension with external commonsense knowledge,

R. Rudinger, S. Istvan, and B. Van Durme, “Knowledgeable reader: Enhancing cloze-style reading comprehension with external commonsense knowledge,” inProceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 1753–1762, 2018

2018
[21]

Xnli: Evaluating cross-lingual sentence representations,

A. Conneau, R. Rinott, G. Lample, A. Williams, S. R. Bowman, H. Schwenk, and V. Stoy- anov, “Xnli: Evaluating cross-lingual sentence representations,” inProceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pp. 2475–2485, 2018

2018
[22]

Attention is all you need,

A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” inAdvances in neural information processing systems, pp. 5998–6008, 2017

2017
[23]

High-precision medical speech recognition through synthetic data and semantic correction: United-medasr,

S. Banerjee, A. Agarwal, and P. Ghosh, “High-precision medical speech recognition through synthetic data and semantic correction: United-medasr,” 2024

2024
[24]

Bart: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension,

M. Lewis, Y. Liu, N. Goyal, M. Ghazvininejad, A. Mohamed, O. Levy, V. Stoyanov, and L. Zettle- moyer, “Bart: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension,” 2019

2019
[25]

Smart: Robust and efficient fine- tuning for pre-trained natural language models through principled regularized optimization,

H. Jiang, P. He, W. Chen, X. Liu, J. Gao, and T. Zhao, “Smart: Robust and efficient fine- tuning for pre-trained natural language models through principled regularized optimization,” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, As...

2020
[26]

Palm 2 technical report,

R. Anil, A. M. Dai, O. Firat, M. Johnson, D. Lepikhin, A. Passos, S. Shakeri, E. Taropa, P. Bailey, Z. Chen, E. Chu, J. H. Clark, L. E. Shafey, Y. Huang, K. Meier-Hellstern, G. Mishra, E. Moreira, M. Omernick, K. Robinson, S. Ruder, Y. Tay, K. Xiao, Y. Xu, Y. Zhang, G. H. Abre...

2023
[27]

Claude 3 model card,

Anthropic, “Claude 3 model card,” tech. rep., Anthropic, 2023

2023
[28]

Gpt-4 technical report,

OpenAI, J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkat, R. Avila, I. Babuschkin, S. Balaji, V. Balcom, P. Bal- tescu, H. Bao, M. Bavarian, J. Belgum, I. Bello, J. Berdine, G. Bernadett-Shapiro, C. Berner, ...

2024
[29]

Scitail: A textual entailment dataset from science question answering,

T. Khot, A. Sabharwal, and P. Clark, “Scitail: A textual entailment dataset from science question answering,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 32, 2018

2018
[30]

Multi-task deep neural networks for natural language un- derstanding,

P. Liu, X. Qiu, and X. Huang, “Multi-task deep neural networks for natural language un- derstanding,” inProceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp. 4487–4496, 2019

2019
[31]

Anli: A new benchmark for natural language understanding,

A. Williams, N. Nangia, and S. R. Bowman, “Anli: A new benchmark for natural language understanding,” inarXiv preprint arXiv:2004.07828, 2020

2004
[32]

e-snli: Natural language inference with natural language explanations,

O.-M. Camburu, T. Rocktäschel, T. Lukasiewicz, and P. Blunsom, “e-snli: Natural language inference with natural language explanations,” inAdvances in Neural Information Processing Systems, pp. 9539–9549, 2018

2018
[33]

Scaling instruction-finetuned language models,

H. W. Chung, L. Hou, S. Longpre, B. Zoph, Y. Tay, W. Fedus, Y. Li, X. Wang, M. Dehghani, S. Brahma, A. Webson, S. S. Gu, Z. Dai, M. Suzgun, X. Chen, A. Chowdhery, A. Castro-Ros, M. Pellat, K. Robinson, D. Valter, S. Narang, G. Mishra, A. Yu, V. Zhao, Y. Huang, A. Dai, H. Yu, S...

2022
[34]

e-snli: Natural language inference with natural language explanations,

O.-M. Camburu, T. Rocktäschel, T. Lukasiewicz, and P. Blunsom, “e-snli: Natural language inference with natural language explanations,” 2018. 14

2018

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

A large annotated corpus for learning natural language inference,

S. R. Bowman, G. Angeli, C. Potts, and C. D. Manning, “A large annotated corpus for learning natural language inference,” inProceedings of the 2015 Conference on Empirical Methods in Natural Language Processing(L. Màrquez, C. Callison-Burch, and J. Su, eds.), (Lisbon, Portugal), pp. 632–642, Association for Computational Linguistics, Sept. 2015

work page 2015

[2] [2]

A decomposable attention model for natural language inference,

A. Parikh, O. Täckström, D. Das, and J. Uszkoreit, “A decomposable attention model for natural language inference,” inProceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pp. 2249–2255, 2016

work page 2016

[3] [3]

Enhanced lstm for natural language inference,

Q. Chen, X. Zhu, Z. Ling, S. Wei, H. Jiang, and D. Inkpen, “Enhanced lstm for natural language inference,” inProceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 1657–1668, 2017

work page 2017

[4] [4]

Improving language understanding by generative pre-training,

A. Radford, K. Narasimhan, T. Salimans, and I. Sutskever, “Improving language understanding by generative pre-training,” 2018. OpenAI preprint

work page 2018

[5] [5]

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,” 2019

2019

[6] [6]

Xlnet: General- ized autoregressive pretraining for language understanding,

Z. Yang, Z. Dai, Y. Yang, J. Carbonell, R. R. Salakhutdinov, and Q. V. Le, “Xlnet: General- ized autoregressive pretraining for language understanding,”Advances in neural information processing systems, vol. 32, 2019

work page 2019

[7] [7]

Roberta: A robustly optimized bert pretraining approach,

Y. Liu, M. Ott, N. Goyal, J. Du, M. Joshi, D. Chen, O. Levy, M. Lewis, L. Zettlemoyer, and V. Stoyanov, “Roberta: A robustly optimized bert pretraining approach,” inarXiv preprint arXiv:1907.11692, 2019

arXiv 1907

[8] [8]

Albert: A lite bert for self-supervised learning of language representations,

Z. Lan, M. Chen, S. Goodman, K. Gimpel, P. Sharma, and R. Soricut, “Albert: A lite bert for self-supervised learning of language representations,” 2020

work page 2020