Causal Reasoning on e-CARE OOD (held-out)
82.3P0 ScoreQwen2.5-14B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen2.5-14BAdapter type=Vernier, QLoRA=true2026.06 | 82.3 | 61 | 21.3 | -1 | |
| Qwen2.5-14BAdapter type=Base, QLoRA=true2026.06 | 82 | 60.5 | 21.5 | — | |
| Qwen2.5-32BAdapter type=Base, QLoRA=true2026.06 | 82 | 62.5 | 19.5 | — | |
| Qwen2.5-7BAdapter type=Base2026.06 | 81.5 | 58.5 | 23 | — | |
| Qwen2.5-32BAdapter type=Vernier, QLoRA=true2026.06 | 81.5 | 59.5 | 22 | 13 | |
| Qwen2.5-7BAdapter type=Vernier2026.06 | 81 | 62.2 | 18.8 | -18 | |
| Phi-3-mini (3.8B)Adapter type=Base2026.06 | 79.5 | 59 | 20.5 | — | |
| Llama-3.1-8BAdapter type=Vernier2026.06 | 79.3 | 61.5 | 17.8 | -26 | |
| Llama-3.1-8BAdapter type=Base2026.06 | 79 | 55 | 24 | — | |
| Mistral-7BAdapter type=Base2026.06 | 79 | 55.5 | 23.5 | — | |
| Phi-3-mini (3.8B)Adapter type=Vernier2026.06 | 78.5 | 60 | 18.5 | -10 | |
| Mistral-7BAdapter type=Vernier, recipe=default, lr=1e-42026.06 | 77.5 | 58 | 19.5 | -17 | |
| Mistral-7BAdapter type=Vernier, recipe=rescued, lr=5e-6, early stopping step=1502026.06 | 77 | 56.5 | 20.5 | -13 |