Hallucination Detection on CoQA (AUROC, F1)
0.98AUROCPCNET
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PCNETModel=Llama-3.1-8B2026.05 | 0.98 | 0.98 | |
| PCNETModel=Mistral-7B-v0.32026.05 | 0.97 | 0.96 | |
| HaloScopeModel=Llama-3.1-8B2026.05 | 0.97 | 0.97 | |
| PCNETModel=Qwen3-4B2026.05 | 0.96 | 0.96 | |
| PCNETModel=Llama-3.2-1B2026.05 | 0.95 | 0.95 | |
| HaloScopeModel=Mistral-7B-v0.32026.05 | 0.88 | 0.88 | |
| HaloScopeModel=Llama-3.2-1B2026.05 | 0.85 | 0.85 | |
| SEPModel=Llama-3.2-1B2026.05 | 0.77 | 0.77 | |
| SEPModel=Mistral-7B-v0.32026.05 | 0.69 | 0.72 | |
| SEPModel=Qwen3-4B2026.05 | 0.65 | 0.73 | |
| HaloScopeModel=Qwen3-4B2026.05 | 0.63 | 0.73 | |
| AutoFactModel=Llama-3.2-1B2026.05 | 0.56 | 0.67 | |
| Token NLLModel=Mistral-7B-v0.32026.05 | 0.56 | 0.67 | |
| AutoFactModel=Mistral-7B-v0.32026.05 | 0.56 | 0.67 | |
| AutoFactModel=Qwen3-4B2026.05 | 0.56 | 0.67 | |
| AutoFactModel=Llama-3.1-8B2026.05 | 0.56 | 0.67 | |
| Token NLLModel=Llama-3.2-1B2026.05 | 0.55 | 0.67 | |
| Token NLLModel=Qwen3-4B2026.05 | 0.55 | 0.67 | |
| Token NLLModel=Llama-3.1-8B2026.05 | 0.55 | 0.67 | |
| SEPModel=Llama-3.1-8B2026.05 | 0.33 | 0.67 |