Natural Language Inference on CB (Accuracy and Macro-F1)
87.5AccuracyQwen1.5-7B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen1.5-7BFine-tuning protocol=LoRA2026.02 | 87.5 | 78.57 | |
| Krause-Qwen1.5-7BWindow size=48, top-k=242026.02 | 87.5 | 78.62 | |
| Krause-Qwen1.5-7BWindow size=32, top-k=162026.02 | 85.71 | 70.09 | |
| Krause-Qwen1.5-7BWindow size=18, top-k=122026.02 | 85.71 | 70.06 | |
| Qwen1.5-7BFine-tuning protocol=None2026.02 | 76.79 | 53.73 | |
| Krause-Llama3-8B2026.02 | 64.29 | 48.04 | |
| Llama3-8Bfine-tuning=LoRA2026.02 | 60.71 | 47.81 | |
| Llama3-8B2026.02 | 41.07 | 19.41 |