Truthful Question Answering on TruthfulQA (817 samples full)
27.78MC1 ScoreDOLA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DOLABackbone=Qwen2.5-7B, Configuration=dynamic2026.06 | 27.78 | 48.22 | |
| CAABackbone=Qwen2.5-7B, Configuration=top_k=10, coeff=5.02026.06 | 25.58 | 43.38 | |
| DOLABackbone=Qwen2.5-7B, Configuration=mid (static)2026.06 | 24.48 | 45.55 | |
| ITIBackbone=Qwen2.5-7B, Configuration=top_k=10, λ=32026.06 | 24.36 | 41.78 | |
| DOLABackbone=Qwen2.5-7B, Configuration=early (static)2026.06 | 23.26 | 43.71 | |
| BaselineBackbone=Qwen2.5-7B, Configuration=—2026.06 | 22.15 | 39.21 | |
| ITIBackbone=Qwen2.5-7B, Configuration=top_k=3, λ=1–32026.06 | 22.15 | 39.21 | |
| CAABackbone=Qwen2.5-7B, Configuration=top_k=3, all coeff2026.06 | 22.15 | 39.21 | |
| ITIBackbone=Qwen2.5-7B, Configuration=top_k=5, λ=1–32026.06 | 22.03 | 39.26 | |
| CAABackbone=Qwen2.5-7B, Configuration=top_k=5, all coeff2026.06 | 22.03 | 39.22 |