Paraphrase Detection on QQP (test)
95.3AccuracyMoE-DiffuSeq
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MoE-DiffuSeq2025.12 | 95.3 | — | — | — | — | |
| Longformer2025.12 | 92.3 | — | — | — | — | |
| DiffuSeq2025.12 | 91.7 | — | — | — | — | |
| FLoRGBase Model=RoBERTa-large2026.02 | 91.52 | — | — | — | — | |
| FedSA-LoRABase Model=RoBERTa-large2026.02 | 90.01 | — | — | — | — | |
| FLoRGBase Model=OPT-125M2026.02 | 89.72 | — | — | — | — | |
| FeDeRABase Model=RoBERTa-large2026.02 | 89.3 | — | — | — | — | |
| FFA-LoRABase Model=RoBERTa-large2026.02 | 88.51 | — | — | — | — | |
| FedSA-LoRABase Model=OPT-125M2026.02 | 87.92 | — | — | — | — | |
| FedITBase Model=RoBERTa-large2026.02 | 87.89 | — | — | — | — | |
| FeDeRABase Model=OPT-125M2026.02 | 87.09 | — | — | — | — | |
| Parallelized LMUParam.=12012021.02 | 86.95 | 85.36 | — | — | — | |
| FFA-LoRABase Model=OPT-125M2026.02 | 86.3 | — | — | — | — | |
| RoBERTaTraining Mode=In-distribution2022.10 | 84.5 | — | 84.4 | — | — | |
| GAPXTraining Mode=In-distribution2022.10 | 84.5 | — | 84.4 | — | — | |
| FedITBase Model=OPT-125M2026.02 | 84.18 | — | — | — | — | |
| BARTTraining Mode=In-distribution2022.10 | 82.8 | — | 82.6 | — | — | |
| BERTTraining Mode=In-distribution2022.10 | 82.6 | — | 82.5 | — | — | |
| LSTMParam.=-/800k2021.02 | 82.58 | 81.4 | — | — | — | |
| BERT+EPTraining Mode=In-distribution2022.10 | 81.7 | — | 81.6 | — | — | |
| IDPTraining Mode=In-distribution2022.10 | 79.1 | — | 79 | — | — | |
| RETROPROMPTShot=16-shot2022.05 | 74 | — | — | — | — | |
| RETROPROMPTshot=16-shot, domain_role=Source2025.12 | 74 | — | — | — | — | |
| OODPTraining Mode=In-distribution2022.10 | 73.2 | — | 65.3 | — | — | |
| BARTTraining Mode=Transfer (Low Shift)2022.10 | 72.7 | — | 68.6 | — | — | |
| KPTShot=16-shot2022.05 | 71.6 | — | — | — | — | |
| KPTshot=16-shot, domain_role=Source2025.12 | 71.6 | — | — | — | — | |
| GAPTraining Mode=In-distribution2022.10 | 71 | — | 68.8 | — | — | |
| Final-exitOffloading cost=5λ2023.09 | 71 | — | — | 436.6 | — | |
| RoBERTaTraining Mode=Transfer (Low Shift)2022.10 | 68.8 | — | 66.5 | — | — | |
| BERTTraining Mode=Transfer (Low Shift)2022.10 | 68.3 | — | 68 | — | — | |
| LM-BFF (D-demo)Shot=16-shot2022.05 | 68.2 | — | — | — | — | |
| LM-BFF (D-demo)shot=16-shot, domain_role=Source2025.12 | 68.2 | — | — | — | — | |
| IDPTraining Mode=Transfer (Low Shift)2022.10 | 66.9 | — | 65.2 | — | — | |
| OODPTraining Mode=Transfer (Low Shift)2022.10 | 65.5 | — | 62.2 | — | — | |
| LM-BFF (man)Shot=16-shot2022.05 | 65.4 | — | — | — | — | |
| LM-BFF (man)shot=16-shot, domain_role=Source2025.12 | 65.4 | — | — | — | — | |
| GAPTraining Mode=Transfer (Low Shift)2022.10 | 65 | — | 62.2 | — | — | |
| BiLSTMTraining Mode=In-distribution2022.10 | 63.6 | — | 61.6 | — | — | |
| GAPXTraining Mode=Transfer (Low Shift)2022.10 | 63.4 | — | 62.3 | — | — | |
| BERT+EPTraining Mode=Transfer (Low Shift)2022.10 | 62.7 | — | 58.6 | — | — | |
| FTShot=16-shot2022.05 | 60.7 | — | — | — | — | |
| FTshot=16-shot, domain_role=Source2025.12 | 60.7 | — | — | — | — | |
| BOWTraining Mode=In-distribution2022.10 | 57.8 | — | 51.3 | — | — | |
| BiLSTMTraining Mode=Transfer (Low Shift)2022.10 | 52.9 | — | 51.7 | — | — | |
| BOWTraining Mode=Transfer (Low Shift)2022.10 | 49.7 | — | 41.7 | — | — | |
| SplitEE-SOffloading cost=5λ2023.09 | 0.1 | — | — | -55.1 | — | |
| Random-exitOffloading cost=5λ2023.09 | -0.1 | — | — | -14.8 | — | |
| SplitEEOffloading cost=5λ2023.09 | -0.1 | — | — | -59.1 | — | |
| ElasticBERTOffloading cost=5λ2023.09 | -0.2 | — | — | -57.9 | — | |
| DeeBERTOffloading cost=5λ2023.09 | -6.7 | — | — | -50.1 | — | |
| ABEXNumber of training samples=1002024.06 | — | — | — | — | 72.13 | |
| ABEXNumber of training samples=2002024.06 | — | — | — | — | 74.32 | |
| ABEXNumber of training samples=5002024.06 | — | — | — | — | 76.53 | |
| ABEXNumber of training samples=10002024.06 | — | — | — | — | 76.81 | |
| ABEX-AbsNumber of training samples=1002024.06 | — | — | — | — | 68.31 | |
| ABEX-AbsNumber of training samples=2002024.06 | — | — | — | — | 70.44 | |
| ABEX-AbsNumber of training samples=5002024.06 | — | — | — | — | 72.3 | |
| ABEX-AbsNumber of training samples=10002024.06 | — | — | — | — | 73.08 | |
| ABEX-stage-1Number of training samples=1002024.06 | — | — | — | — | 71.6 | |
| ABEX-stage-1Number of training samples=2002024.06 | — | — | — | — | 74.02 | |
| ABEX-stage-1Number of training samples=5002024.06 | — | — | — | — | 76.49 | |
| ABEX-stage-1Number of training samples=10002024.06 | — | — | — | — | 76.73 | |
| ABEX-stage-2Number of training samples=1002024.06 | — | — | — | — | 70.24 | |
| ABEX-stage-2Number of training samples=2002024.06 | — | — | — | — | 71.68 | |
| ABEX-stage-2Number of training samples=5002024.06 | — | — | — | — | 74.57 | |
| ABEX-stage-2Number of training samples=10002024.06 | — | — | — | — | 74.89 | |
| AEDANumber of training samples=1002024.06 | — | — | — | — | 69.45 | |
| AEDANumber of training samples=2002024.06 | — | — | — | — | 68.81 | |
| AEDANumber of training samples=5002024.06 | — | — | — | — | 72.54 | |
| AEDANumber of training samples=10002024.06 | — | — | — | — | 76.32 | |
| AMR-DANumber of training samples=1002024.06 | — | — | — | — | 69.58 | |
| AMR-DANumber of training samples=2002024.06 | — | — | — | — | 70.63 | |
| AMR-DANumber of training samples=5002024.06 | — | — | — | — | 72.31 | |
| AMR-DANumber of training samples=10002024.06 | — | — | — | — | 73.66 | |
| BackTransNumber of training samples=1002024.06 | — | — | — | — | 67.21 | |
| BackTransNumber of training samples=2002024.06 | — | — | — | — | 69.44 | |
| BackTransNumber of training samples=5002024.06 | — | — | — | — | 71.43 | |
| BackTransNumber of training samples=10002024.06 | — | — | — | — | 72.34 | |
| DARTBackbone=RoBERTa-large, Training Samples (K)=16, Evaluation Protocol=Few-shot fine-tuning2021.08 | — | 67.8 | — | — | — | |
| EDANumber of training samples=1002024.06 | — | — | — | — | 69.22 | |
| EDANumber of training samples=2002024.06 | — | — | — | — | 69.51 | |
| EDANumber of training samples=5002024.06 | — | — | — | — | 70.64 | |
| EDANumber of training samples=10002024.06 | — | — | — | — | 73.02 | |
| Fine-tuningBackbone=RoBERTa-large, Training Samples (K)=16, Evaluation Protocol=Few-shot fine-tuning2021.08 | — | 60.7 | — | — | — | |
| Fine-tuning (full)Backbone=RoBERTa-large, Training Samples (K)=full, Evaluation Protocol=Full fine-tuning2021.08 | — | 81.7 | — | — | — | |
| Gold-onlyNumber of training samples=1002024.06 | — | — | — | — | 69.23 | |
| Gold-onlyNumber of training samples=2002024.06 | — | — | — | — | 72 | |
| Gold-onlyNumber of training samples=5002024.06 | — | — | — | — | 75.27 | |
| Gold-onlyNumber of training samples=10002024.06 | — | — | — | — | 76.15 | |
| GPT-3 in-context learningBackbone=RoBERTa-large, Training Samples (K)=16, Evaluation Protocol=In-context learning2021.08 | — | 36.1 | — | — | — | |
| LLAMA-213BNumber of training samples=1002024.06 | — | — | — | — | 70.35 | |
| LLAMA-213BNumber of training samples=2002024.06 | — | — | — | — | 73.57 | |
| LLAMA-213BNumber of training samples=5002024.06 | — | — | — | — | 74.39 | |
| LLAMA-213BNumber of training samples=10002024.06 | — | — | — | — | 74.81 | |
| LM-BFFBackbone=RoBERTa-large, Training Samples (K)=16, Evaluation Protocol=Few-shot fine-tuning2021.08 | — | 67 | — | — | — | |
| MajorityBackbone=RoBERTa-large, Training Samples (K)=0, Evaluation Protocol=Majority class2021.08 | — | 0 | — | — | — | |
| P-TuningBackbone=RoBERTa-large, Training Samples (K)=16, Evaluation Protocol=Few-shot fine-tuning2021.08 | — | 65.6 | — | — | — | |
| Prompt-based zero-shotBackbone=RoBERTa-large, Training Samples (K)=0, Evaluation Protocol=Zero-shot2021.08 | — | 49.7 | — | — | — | |
| SSMBANumber of training samples=1002024.06 | — | — | — | — | 66.51 |