List your company

Hallucination Detection on Winogrande (test)

52.48AuROCSpilled ΔE
47.945649.122850.351.4772Feb 21, 2026

Evaluation results

MethodLinks
Spilled ΔEModel=Gemma-Instruct 1B, Pool=Min2026.0252.48
Spilled ΔEsModel=Gemma-Instruct 1B, Pool=Max2026.0252.27
Marginal EmModel=Gemma-Instruct 4B, Pool=Max2026.0251.49
Spilled ΔEsModel=Gemma-Instruct 4B, Pool=Max2026.0250.59
Logit EℓModel=Gemma-Instruct 1B, Pool=Max2026.0250.5
Marginal EmModel=Gemma-Instruct 1B, Pool=Max2026.0250.23
Spilled ΔEModel=Gemma-Instruct 4B, Pool=Min2026.0250.18
Marginal EmModel=Gemma-Instruct 1B, Pool=Min2026.0249.84
Logit EℓModel=Gemma-Instruct 4B, Pool=Max2026.0249.77
Marginal EmModel=Gemma-Instruct 4B, Pool=Min2026.0248.12