Natural Language Inference on GLUE
91.45MNLI AccuracyLoRA-Mixer
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| LoRA-Mixer2025.06 | 91.45 | 92.04 | — | — | — | — | — | |
| MixLoRA2025.06 | 91.27 | 91.73 | — | — | — | — | — | |
| MOLE2025.06 | 91.04 | 91.48 | — | — | — | — | — | |
| LoRA2025.06 | 90.87 | 91.34 | — | — | — | — | — | |
| LoRAHub2025.06 | 90.56 | 90.44 | — | — | — | — | — | |
| Base2025.06 | 89.95 | 90.36 | — | — | — | — | — | |
| θB fine-tuneSource Architecture=T5-3B, Target Architecture=T5-Large, Alignment Batches (B)=–, Mode=Fine-tuned2026.02 | 89.8 | 94.27 | 83.59 | 95.86 | 91.83 | 91.09 | 48.72 | |
| X-FMModel size=base-size2023.01 | 87.7 | — | — | — | — | — | — | |
| RoBERTaModel size=base-size2023.01 | 87.6 | — | — | — | — | — | — | |
| mPLUG-2Model size=base-size2023.01 | 87.6 | — | — | — | — | — | — | |
| UNIMO-2Model size=base-size2023.01 | 87.5 | — | — | — | — | — | — | |
| OFAModel size=base-size2023.01 | 84.3 | — | — | — | — | — | — | |
| SimVLMModel size=base-size2023.01 | 83.4 | — | — | — | — | — | — | |
| DaVinciModel size=base-size2023.01 | 83.1 | — | — | — | — | — | — | |
| Uni-Perceiver-MoEModel size=base-size2023.01 | 81.5 | — | — | — | — | — | — | |
| FLAVAModel size=base-size2023.01 | 80.3 | — | — | — | — | — | — | |
| THESEUSSource Architecture=T5-3B, Target Architecture=T5-Large, Alignment Batches (B)=202026.02 | 57.01 | 50.56 | 49.46 | 50.77 | 57.24 | 53.01 | 10.64 | |
| THESEUSSource Architecture=T5-3B, Target Architecture=T5-Large, Alignment Batches (B)=1002026.02 | 55.66 | 59.49 | 50.98 | 63.11 | 64.89 | 58.83 | 16.46 | |
| THESEUSSource Architecture=T5-3B, Target Architecture=T5-Large, Alignment Batches (B)=502026.02 | 54.52 | 56.31 | 50.12 | 53.45 | 60.42 | 54.96 | 12.59 | |
| θBSource Architecture=T5-3B, Target Architecture=T5-Large, Alignment Batches (B)=–, Mode=Zero-shot2026.02 | 32.23 | 48.05 | 48.38 | 50.77 | 32.44 | 42.37 | 0 |