Selective Prediction on NCQA (test)
99.1PRRLlama-3.1-8B
Evaluation results
| Method | Links | |
|---|---|---|
| Llama-3.1-8BFeatures=all UQ, Strategy=oracle gated, Supervision=AlignScore2026.06 | 99.1 | |
| Llama-3.1-8BFeatures=all UQ, Strategy=oracle selective, Supervision=AlignScore2026.06 | 99.1 | |
| Spectrum-Qwen3-14BFeatures=all UQ, Strategy=baseline, Supervision=AlignScore2026.06 | 69.5 | |
| A^2Search-7BFeatures=SAR, Strategy=baseline, Supervision=AlignScore2026.06 | 68.3 | |
| A^2Search-7BFeatures=SAR, Strategy=latent gated, Supervision=AlignScore2026.06 | 67 | |
| Qwen3-14BFeatures=MSP, Strategy=latent selective, Supervision=AlignScore2026.06 | 64.5 | |
| A^2Search-7BFeatures=MSP, Strategy=latent gated, Supervision=AlignScore2026.06 | 63.8 | |
| A^2Search-7BFeatures=Semantic energy, Strategy=latent selective, Supervision=AlignScore2026.06 | 63 | |
| Qwen3-14BFeatures=CoCoA, Strategy=latent gated, Supervision=AlignScore2026.06 | 60.6 | |
| Spectrum-Qwen3-14BFeatures=all UQ, Strategy=latent selective, Supervision=AlignScore2026.06 | 59.9 | |
| Qwen3-14BFeatures=all UQ, Strategy=latent gated, Supervision=AlignScore2026.06 | 57.7 | |
| Qwen3-14BFeatures=all UQ, Strategy=latent selective, Supervision=AlignScore2026.06 | 53.4 | |
| Llama-3.1-8BFeatures=SAR, Strategy=latent gated, Supervision=AlignScore2026.06 | 50.7 | |
| Llama-3.1-8BFeatures=MI, Strategy=latent gated, Supervision=AlignScore2026.06 | 50.2 | |
| Qwen3-14BFeatures=all UQ, Strategy=baseline, Supervision=AlignScore2026.06 | 49.9 | |
| Llama-3.1-8BFeatures=all UQ, Strategy=latent selective, Supervision=AlignScore2026.06 | 44.2 | |
| Spectrum-Llama-3.1-8BFeatures=MI, Strategy=latent gated, Supervision=AlignScore2026.06 | 43.3 | |
| Llama-3.1-8BFeatures=all UQ, Strategy=latent gated, Supervision=AlignScore2026.06 | 36.9 | |
| Llama-3.1-8BFeatures=all UQ, Strategy=baseline, Supervision=AlignScore2026.06 | 33 |