Natural Language Inference on SNLI (test)
94.7AccuracyUnitedSynT5
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| UnitedSynT5Year=2024, Parameters=3B2024.12 | 94.7 | — | — | |
| UnitedSynT5Year=2024, Parameters=335M2024.12 | 93.5 | — | — | |
| ALUM_ROBERTA-LARGE-SMARTBackbone=RoBERTa-Large, Adversarial Pre-training=ALUM, Adversarial Fine-tuning=SMART2020.04 | 93.4 | — | — | |
| EFLYear=20212024.12 | 93.1 | — | — | |
| ALUM_ROBERTA-LARGEBackbone=RoBERTa-Large, Adversarial Pre-training=ALUM2020.04 | 93 | — | — | |
| Fine-tuning (full)Backbone=RoBERTa-large, Training Samples (K)=full, Evaluation Protocol=Full fine-tuning2021.08 | 92.6 | — | — | |
| HDGCNTotal parameters=1.02x, Data used (%)=10%2021.06 | 92.3 | — | — | |
| CA-MTLBackbone=RoBERTa-Large2020.09 | 92.1 | — | — | |
| SemBERT_WWMscale=Large, pre-training=Whole Word Masking, evaluation protocol=fine-tuned2019.09 | 91.9 | — | — | |
| SemBERT2020.09 | 91.9 | — | — | |
| DeBERTa-v3-largeYear=2021, Parameters=350M2024.12 | 91.9 | — | — | |
| RoBERTa-LargeParameters (approx.)=355M, Fine-tuning=true2026.03 | 91.83 | — | — | |
| ALBERT-xxlargeYear=2020, Parameters=223M2024.12 | 91.8 | — | — | |
| MT-DNN-SMART_LARGE_V0backbone=BERT-Large, architecture=MT-DNN, framework=SMART, version=v02019.11 | 91.7 | — | — | |
| StructBERTLarge2019.08 | 91.7 | — | — | |
| MT-DNNOptimization=SMART2020.09 | 91.7 | — | — | |
| RoBERTa-largeYear=2019, Parameters=355M2024.12 | 91.7 | — | — | |
| MT-DNN_LARGEbackbone=BERT-Large, architecture=MT-DNN2019.11 | 91.6 | — | — | |
| MT-DNNModel Architecture=LARGE2019.01 | 91.6 | — | — | |
| MT-DNNevaluation protocol=fine-tuned2019.09 | 91.6 | — | — | |
| SemBERT_LARGEscale=Large, evaluation protocol=fine-tuned2019.09 | 91.6 | — | — | |
| BERT_WWMscale=Large, pre-training=Whole Word Masking, evaluation protocol=fine-tuned2019.09 | 91.6 | — | — | |
| MT-DNN_LARGEBackbone=MT-DNN-Large2020.04 | 91.6 | — | — | |
| MT-DNN2020.09 | 91.6 | — | — | |
| XLNet-largeYear=2019, Parameters=340M2024.12 | 91.6 | — | — | |
| MT-DNN-SMART_BASEbackbone=BERT-Base, architecture=MT-DNN, framework=SMART2019.11 | 91.5 | — | — | |
| MT-DNN-SMART_BASE_V0backbone=BERT-Base, architecture=MT-DNN, framework=SMART, version=v02019.11 | 91.4 | — | — | |
| BERT_LARGE + SRLbackbone=BERT-Large, extra_features=SRL2019.11 | 91.3 | — | — | |
| SJRC2019.08 | 91.3 | — | — | |
| Baseline (BERT_LARGE) + SRLbackbone=BERT-Large, framework=SRL enhancement framework, encoding_type=SRL based model2018.09 | 91.3 | — | — | |
| SJRC2019.09 | 91.3 | — | — | |
| CLINEmode=fine-tuned2021.07 | 91.3 | — | — | |
| StandardModel=RoBERTa-base2021.12 | 91.3 | — | — | |
| MT-DNN_BASEbackbone=BERT-Base, architecture=MT-DNN2019.11 | 91.1 | — | — | |
| SMART_BERT-BASEbackbone=BERT-Base, framework=SMART2019.11 | 91.1 | — | — | |
| MT-DNNModel Architecture=BASE2019.01 | 91.1 | — | — | |
| MT-DNN2019.08 | 91.1 | — | — | |
| MT-DNNencoding_type=joint method, notes=previous state-of-the-art2018.09 | 91.1 | — | — | |
| BERT_LARGEscale=Large, evaluation protocol=fine-tuned2019.09 | 91.1 | — | — | |
| MT-DNNData used (%)=10%2021.06 | 91.1 | — | — | |
| BERT_LARGEbackbone=BERT-Large2019.11 | 91 | — | — | |
| BERTModel Architecture=LARGE2019.01 | 91 | — | — | |
| SemBERT_BASEscale=Base, evaluation protocol=fine-tuned2019.09 | 91 | — | — | |
| BERT_LARGEBackbone=BERT-Large2020.04 | 91 | — | — | |
| ALBERT-LargeParameters (approx.)=17M, Fine-tuning=true2026.03 | 90.85 | — | — | |
| BERT_BASEbackbone=BERT-Base2019.11 | 90.8 | — | — | |
| BERTModel Architecture=BASE2019.01 | 90.8 | — | — | |
| BERT2019.08 | 90.8 | — | — | |
| RoBERTamode=fine-tuned2021.07 | 90.8 | — | — | |
| BERT_BASEscale=Base, evaluation protocol=fine-tuned2019.09 | 90.7 | — | — | |
| BERT teacherRole=Teacher, Architecture=BERT2023.05 | 90.59 | — | — | |
| Baseline (BERT_LARGE)backbone=BERT-Large, encoding_type=SRL based model (Baseline)2018.09 | 90.4 | — | — | |
| BERT-LargeParameters (approx.)=340M, Fine-tuning=true2026.03 | 90.33 | — | — | |
| BERT_BASE + SRLbackbone=BERT-Base, extra_features=SRL2019.11 | 90.3 | — | — | |
| DRCNMethod Category=Joint method, Number of Parameters=6.7m, Ensemble=true2018.05 | 90.1 | — | — | |
| Kim et al.2019.01 | 90.1 | — | — | |
| DRCNParams=53.3M, Ensemble=true2019.08 | 90.1 | — | — | |
| DRCN2019.09 | 90.1 | — | — | |
| BERT-largeYear=2019, Parameters=340M2024.12 | 90.1 | — | — | |
| RoBERTa-largeParameters=355M2025.11 | 90 | — | — | |
| LM-TransformerMethod Category=Joint method, Number of Parameters=85m2018.05 | 89.9 | — | — | |
| GPT2018.04 | 89.9 | — | — | |
| GPT2019.01 | 89.9 | — | — | |
| RE2Params=22.4M, Ensemble=true2019.08 | 89.9 | — | — | |
| GPT2019.08 | 89.9 | — | — | |
| LM-Transformerencoding_type=joint method2018.09 | 89.9 | — | — | |
| GPT2020.04 | 89.9 | — | — | |
| GPTYear=2018, Parameters=117M2024.12 | 89.9 | — | — | |
| BERTmode=fine-tuned2021.07 | 89.8 | — | — | |
| DR-BiLSTMmode=Ensemble, preprocessing=enabled2018.02 | 89.6 | — | — | |
| DMANModel Architecture Class=Ensemble2019.07 | 89.6 | — | — | |
| Baseline (BERT_BASE) + SRLbackbone=BERT-Base, framework=SRL enhancement framework, encoding_type=SRL based model2018.09 | 89.6 | — | — | |
| MwANParams=58M, Ensemble=true2019.08 | 89.4 | — | — | |
| ESIM+ELMoEnsemble=true2018.02 | 89.3 | — | — | |
| DR-BiLSTMMethod Category=Joint method, Number of Parameters=7.5m, Ensemble=true2018.05 | 89.3 | — | — | |
| CAFEMethod Category=Joint method, Number of Parameters=4.7m, Ensemble=true2018.05 | 89.3 | — | — | |
| ESIM+ELMoMethod Category=Joint method, Number of Parameters=8.0m, Ensemble=true2018.05 | 89.3 | — | — | |
| DR-BiLSTMmode=Ensemble2018.02 | 89.3 | — | — | |
| CAFEEmbedding Dimension=300D, Model Architecture Class=Ensemble2019.07 | 89.3 | — | — | |
| CAFEParams=17.5M, Ensemble=true2019.08 | 89.3 | — | — | |
| Baseline (BERT_BASE)backbone=BERT-Base, encoding_type=SRL based model (Baseline)2018.09 | 89.2 | — | — | |
| StandardModel=BERT-base-uncased2021.12 | 89.2 | — | — | |
| StandardModel=BERT-base-uncased2021.12 | 89.2 | — | — | |
| ESIM with SuBiLSTMBase Model=ESIM, Ensemble=5 models2018.05 | 89.1 | — | — | |
| ESIM with SuBiLSTM-TiedBase Model=ESIM, Ensemble=5 models2018.05 | 89.1 | — | — | |
| KIMMethod Category=Joint method, Number of Parameters=4.3m, Ensemble=true2018.05 | 89.1 | — | — | |
| KIMModel Architecture Class=Ensemble2019.07 | 89.1 | — | — | |
| ESIM Baseline (ELMo) + SRLembedding=ELMo, framework=SRL enhancement framework, encoding_type=SRL based model2018.09 | 89.1 | — | — | |
| KIM Ensemble2018.04 | 89 | — | — | |
| DIINensemble=true2017.09 | 88.9 | — | — | |
| DIINEnsemble=true2018.02 | 88.9 | — | — | |
| DIINMethod Category=Joint method, Number of Parameters=4.4m, Ensemble=true2018.05 | 88.9 | — | — | |
| DRCNMethod Category=Joint method, Number of Parameters=6.7m, Ensemble=false2018.05 | 88.9 | — | — | |
| Gong et al.mode=Ensemble2018.02 | 88.9 | — | — | |
| DR-BiLSTMmode=Single, preprocessing=enabled2018.02 | 88.9 | — | — | |
| DIINModel Architecture Class=Ensemble2019.07 | 88.9 | — | — | |
| DRCNParams=6.7M, Ensemble=false2019.08 | 88.9 | — | — | |
| RE2Params=2.8M, Ensemble=false2019.08 | 88.9 | — | — | |
| DIINParams=17M, Ensemble=true2019.08 | 88.9 | — | — | |
| DCRCNencoding_type=joint method2018.09 | 88.9 | — | — |