Multilingual Natural Language Inference on XNLI (test)
86Accuracy (EN)Hybrid
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| HybridBackbone=Wizard-Vicuna-30B, 32-shot=true2025.12 | 86 | 83.4 | 78.9 | 82.8 | 7.1 | 1.7 | |
| LoRA-GABackbone=Wizard-Vicuna-30B, 32-shot=true2025.12 | 85.4 | 82.7 | 78 | 82 | 7.4 | 1.9 | |
| BOFTBackbone=Wizard-Vicuna-30B, 32-shot=true2025.12 | 85.1 | 82.4 | 77.8 | 81.8 | 7.3 | 2 | |
| LoRABackbone=Wizard-Vicuna-30B, 32-shot=true2025.12 | 84.7 | 81.9 | 77.2 | 81.3 | 7.5 | 2.2 | |
| HybridBackbone=BloomZ-7B1, 32-shot=true2025.12 | 82.1 | 79.3 | 75.2 | 78.9 | 6.9 | 2.1 | |
| LoRA-GABackbone=BloomZ-7B1, 32-shot=true2025.12 | 81.4 | 78.6 | 74.3 | 78.1 | 7.1 | 2.4 | |
| BOFTBackbone=BloomZ-7B1, 32-shot=true2025.12 | 81.2 | 78.4 | 74.1 | 77.9 | 7.1 | 2.5 | |
| LoRABackbone=BloomZ-7B1, 32-shot=true2025.12 | 80.6 | 77.9 | 73.5 | 77.3 | 7.1 | 2.8 |