Hallucination Detection on Winobias (test)
54.43AUROCLogit Eℓ
Evaluation Results
| Method | Links | |
|---|---|---|
| Logit EℓModel=Gemma-Instruct 4B, Pool=Max2026.02 | 54.43 | |
| Spilled ΔEsModel=Gemma-Instruct 4B, Pool=Max2026.02 | 53.53 | |
| Logit EℓModel=Gemma-Instruct 1B, Pool=Max2026.02 | 53.49 | |
| Spilled ΔEModel=Gemma-Instruct 4B, Pool=Min2026.02 | 53.1 | |
| Marginal EmModel=Gemma-Instruct 4B, Pool=Min2026.02 | 51.91 | |
| Marginal EmModel=Gemma-Instruct 1B, Pool=Min2026.02 | 50.39 | |
| Marginal EmModel=Gemma-Instruct 1B, Pool=Max2026.02 | 49.93 | |
| Spilled ΔEsModel=Gemma-Instruct 1B, Pool=Max2026.02 | 49.91 | |
| Spilled ΔEModel=Gemma-Instruct 1B, Pool=Min2026.02 | 49.77 | |
| Marginal EmModel=Gemma-Instruct 4B, Pool=Max2026.02 | 47.97 |