Question Answering on TruthfulQA (full dataset, Accuracy)
86.8AccuracyPEAR
Evaluation Results
| Method | Links | |
|---|---|---|
| PEARBackbone=Qwen2.5-14B2026.05 | 86.8 | |
| CliqueBackbone=Qwen2.5-14B2026.05 | 85.7 | |
| StarBackbone=Qwen2.5-14B2026.05 | 83.4 | |
| RingBackbone=Qwen2.5-14B2026.05 | 82.7 | |
| PEARBackbone=Llama-3.1-8B2026.05 | 82.4 | |
| ChainBackbone=Qwen2.5-14B2026.05 | 79.5 | |
| CoT-SCBackbone=Qwen2.5-14B2026.05 | 78.5 | |
| CliqueBackbone=Llama-3.1-8B2026.05 | 78.5 | |
| RingBackbone=Llama-3.1-8B2026.05 | 75.7 | |
| ChainBackbone=Llama-3.1-8B2026.05 | 74.5 | |
| StarBackbone=Llama-3.1-8B2026.05 | 72.2 | |
| CoTBackbone=Qwen2.5-14B2026.05 | 70.9 | |
| CoT-SCBackbone=Llama-3.1-8B2026.05 | 67.5 | |
| CoTBackbone=Llama-3.1-8B2026.05 | 65.3 |