List your company

Prompt Selection on Perturbation-100 (test)

76Top-1 HitRisk-Cos
51.0457.526470.48Mar 20, 2026

Evaluation results

MethodLinks
Risk-CosMode=Prefill-only2026.037618.80.6040.5520.80.17
Risk-MarginMode=Prefill-only2026.037632.50.4810.43-3.8-0.02
Risk-EntropyMode=Prefill-only2026.035765.70.3810.33-23.9-0.18
Risk-Loss (NLL)Mode=Prefill-only2026.0352870.3590.31-28.2-0.23