Selective Prediction on TriviaQA (val)
87.9PRRLlama-3.1-8B
Evaluation Results
| Method | Links | |
|---|---|---|
| Llama-3.1-8BFeatures=all UQ, Protocol=latent selective2026.06 | 87.9 | |
| Llama-3.1-8BFeatures=all UQ, Protocol=baseline2026.06 | 87.8 | |
| Llama-3.1-8BFeatures=all UQ, Protocol=latent gated2026.06 | 87.8 | |
| Llama-3.1-8BFeatures=all UQ, Protocol=oracle gated2026.06 | 87.8 | |
| Spectrum-Qwen3-14BFeatures=all UQ, Protocol=latent selective2026.06 | 87.6 | |
| Spectrum-Qwen3-14BFeatures=all UQ, Protocol=latent gated2026.06 | 87.2 | |
| Spectrum-Qwen3-14BFeatures=all UQ, Protocol=baseline2026.06 | 87.1 | |
| Spectrum-Qwen3-14BFeatures=all UQ, Protocol=oracle gated2026.06 | 87.1 | |
| Llama-3.1-8BFeatures=all UQ, Protocol=oracle selective2026.06 | 86.9 | |
| Spectrum-Qwen3-14BFeatures=SAR, Protocol=latent gated2026.06 | 86.5 | |
| Spectrum-Qwen3-14BFeatures=SAR, Protocol=latent selective2026.06 | 86.5 | |
| Spectrum-Qwen3-14BFeatures=all UQ, Protocol=oracle selective2026.06 | 86.4 | |
| Spectrum-Qwen3-14BFeatures=SAR, Protocol=oracle gated2026.06 | 86.3 | |
| Llama-3.1-8BFeatures=SAR, Protocol=latent gated2026.06 | 86.2 | |
| Llama-3.1-8BFeatures=SAR, Protocol=latent selective2026.06 | 86.2 | |
| Spectrum-Qwen3-14BFeatures=SAR, Protocol=baseline2026.06 | 86.2 | |
| Llama-3.1-8BFeatures=SAR, Protocol=baseline2026.06 | 86.1 | |
| Llama-3.1-8BFeatures=SAR, Protocol=oracle gated2026.06 | 86.1 | |
| Spectrum-Llama-3.1-8BFeatures=all UQ, Protocol=latent selective2026.06 | 85.6 | |
| Spectrum-Qwen3-14BFeatures=SAR, Protocol=oracle selective2026.06 | 85.4 | |
| Spectrum-Llama-3.1-8BFeatures=all UQ, Protocol=latent gated2026.06 | 85.3 | |
| Llama-3.1-8BFeatures=SAR, Protocol=oracle selective2026.06 | 85.2 | |
| Spectrum-Llama-3.1-8BFeatures=all UQ, Protocol=oracle gated2026.06 | 85.2 | |
| Llama-3.1-8BFeatures=CoCoA, Protocol=latent gated2026.06 | 85.1 | |
| Llama-3.1-8BFeatures=CoCoA, Protocol=latent selective2026.06 | 85.1 | |
| Spectrum-Llama-3.1-8BFeatures=all UQ, Protocol=baseline2026.06 | 85.1 | |
| Llama-3.1-8BFeatures=CoCoA, Protocol=baseline2026.06 | 85 | |
| Llama-3.1-8BFeatures=CoCoA, Protocol=oracle gated2026.06 | 85 | |
| Spectrum-Qwen3-14BFeatures=Semantic entropy, Protocol=latent selective2026.06 | 85 | |
| Spectrum-Qwen3-14BFeatures=CoCoA, Protocol=latent gated2026.06 | 84.9 | |
| Spectrum-Qwen3-14BFeatures=CoCoA, Protocol=latent selective2026.06 | 84.9 | |
| Spectrum-Qwen3-14BFeatures=Semantic entropy, Protocol=latent gated2026.06 | 84.8 | |
| Llama-3.1-8BFeatures=Semantic entropy, Protocol=latent selective2026.06 | 84.7 | |
| Spectrum-Qwen3-14BFeatures=CoCoA, Protocol=oracle gated2026.06 | 84.7 | |
| Llama-3.1-8BFeatures=Semantic entropy, Protocol=latent gated2026.06 | 84.6 | |
| Spectrum-Qwen3-14BFeatures=CoCoA, Protocol=baseline2026.06 | 84.6 | |
| Spectrum-Qwen3-14BFeatures=Semantic entropy, Protocol=oracle gated2026.06 | 84.6 | |
| Spectrum-Qwen3-14BFeatures=Semantic entropy, Protocol=baseline2026.06 | 84.5 | |
| Llama-3.1-8BFeatures=Semantic entropy, Protocol=oracle gated2026.06 | 84.4 | |
| Spectrum-Llama-3.1-8BFeatures=all UQ, Protocol=oracle selective2026.06 | 84.3 | |
| Llama-3.1-8BFeatures=CoCoA, Protocol=oracle selective2026.06 | 84.2 | |
| Spectrum-Llama-3.1-8BFeatures=SAR, Protocol=latent selective2026.06 | 84.2 | |
| Llama-3.1-8BFeatures=Semantic entropy, Protocol=baseline2026.06 | 84.1 | |
| Spectrum-Qwen3-14BFeatures=Semantic energy, Protocol=latent selective2026.06 | 84.1 | |
| Spectrum-Llama-3.1-8BFeatures=SAR, Protocol=latent gated2026.06 | 84 | |
| Llama-3.1-8BFeatures=Semantic energy, Protocol=latent selective2026.06 | 83.9 | |
| Spectrum-Qwen3-14BFeatures=Semantic energy, Protocol=latent gated2026.06 | 83.9 | |
| Spectrum-Qwen3-14BFeatures=Semantic entropy, Protocol=oracle selective2026.06 | 83.9 | |
| Llama-3.1-8BFeatures=Semantic energy, Protocol=latent gated2026.06 | 83.8 | |
| Spectrum-Qwen3-14BFeatures=CoCoA, Protocol=oracle selective2026.06 | 83.8 | |
| Llama-3.1-8BFeatures=Semantic energy, Protocol=oracle gated2026.06 | 83.6 | |
| Spectrum-Qwen3-14BFeatures=Semantic energy, Protocol=oracle gated2026.06 | 83.6 | |
| Llama-3.1-8BFeatures=Semantic energy, Protocol=baseline2026.06 | 83.5 | |
| Llama-3.1-8BFeatures=Semantic entropy, Protocol=oracle selective2026.06 | 83.5 | |
| Spectrum-Llama-3.1-8BFeatures=SAR, Protocol=baseline2026.06 | 83.5 | |
| Spectrum-Llama-3.1-8BFeatures=SAR, Protocol=oracle gated2026.06 | 83.5 | |
| Spectrum-Qwen3-14BFeatures=Semantic energy, Protocol=baseline2026.06 | 83.5 | |
| Spectrum-Llama-3.1-8BFeatures=Semantic entropy, Protocol=latent selective2026.06 | 83 | |
| Spectrum-Qwen3-14BFeatures=MI, Protocol=latent gated2026.06 | 83 | |
| Llama-3.1-8BFeatures=MI, Protocol=latent selective2026.06 | 82.9 | |
| Spectrum-Qwen3-14BFeatures=Semantic energy, Protocol=oracle selective2026.06 | 82.9 | |
| Spectrum-Qwen3-14BFeatures=MI, Protocol=latent selective2026.06 | 82.9 | |
| Spectrum-Llama-3.1-8BFeatures=Semantic entropy, Protocol=latent gated2026.06 | 82.8 | |
| Llama-3.1-8BFeatures=MI, Protocol=latent gated2026.06 | 82.6 | |
| Spectrum-Llama-3.1-8BFeatures=SAR, Protocol=oracle selective2026.06 | 82.6 | |
| Llama-3.1-8BFeatures=Semantic energy, Protocol=oracle selective2026.06 | 82.5 | |
| Spectrum-Llama-3.1-8BFeatures=Semantic energy, Protocol=latent selective2026.06 | 82.3 | |
| Spectrum-Llama-3.1-8BFeatures=Semantic entropy, Protocol=oracle gated2026.06 | 82.2 | |
| Qwen3-14BFeatures=all UQ, Protocol=latent selective2026.06 | 82.1 | |
| Spectrum-Llama-3.1-8BFeatures=Semantic energy, Protocol=latent gated2026.06 | 82.1 | |
| Spectrum-Llama-3.1-8BFeatures=CoCoA, Protocol=latent selective2026.06 | 81.9 | |
| Spectrum-Llama-3.1-8BFeatures=Semantic entropy, Protocol=baseline2026.06 | 81.9 | |
| Llama-3.1-8BFeatures=MSP, Protocol=latent selective2026.06 | 81.6 | |
| Spectrum-Qwen3-14BFeatures=MI, Protocol=baseline2026.06 | 81.6 | |
| Llama-3.1-8BFeatures=MI, Protocol=oracle gated2026.06 | 81.4 | |
| Spectrum-Llama-3.1-8BFeatures=CoCoA, Protocol=latent gated2026.06 | 81.4 | |
| Spectrum-Llama-3.1-8BFeatures=Semantic energy, Protocol=oracle gated2026.06 | 81.4 | |
| Llama-3.1-8BFeatures=MSP, Protocol=latent gated2026.06 | 81.2 | |
| Spectrum-Llama-3.1-8BFeatures=Semantic energy, Protocol=baseline2026.06 | 81.2 | |
| Spectrum-Llama-3.1-8BFeatures=Semantic entropy, Protocol=oracle selective2026.06 | 81.2 | |
| Spectrum-Qwen3-14BFeatures=MSP, Protocol=latent selective2026.06 | 81.1 | |
| Spectrum-Llama-3.1-8BFeatures=CoCoA, Protocol=baseline2026.06 | 81 | |
| Spectrum-Llama-3.1-8BFeatures=CoCoA, Protocol=oracle gated2026.06 | 81 | |
| Llama-3.1-8BFeatures=MI, Protocol=baseline2026.06 | 80.9 | |
| Llama-3.1-8BFeatures=MSP, Protocol=baseline2026.06 | 80.8 | |
| Llama-3.1-8BFeatures=MSP, Protocol=oracle gated2026.06 | 80.8 | |
| Spectrum-Llama-3.1-8BFeatures=Semantic energy, Protocol=oracle selective2026.06 | 80.5 | |
| Spectrum-Qwen3-14BFeatures=MSP, Protocol=latent gated2026.06 | 80.4 | |
| Qwen3-14BFeatures=all UQ, Protocol=latent gated2026.06 | 80.3 | |
| Llama-3.1-8BFeatures=MI, Protocol=oracle selective2026.06 | 80.2 | |
| Spectrum-Qwen3-14BFeatures=MI, Protocol=oracle selective2026.06 | 80.2 | |
| Spectrum-Qwen3-14BFeatures=MI, Protocol=oracle gated2026.06 | 80.1 | |
| Llama-3.1-8BFeatures=MSP, Protocol=oracle selective2026.06 | 79.8 | |
| Spectrum-Llama-3.1-8BFeatures=CoCoA, Protocol=oracle selective2026.06 | 79.8 | |
| Qwen3-14BFeatures=all UQ, Protocol=oracle gated2026.06 | 79.6 | |
| Spectrum-Llama-3.1-8BFeatures=MI, Protocol=latent selective2026.06 | 79.6 | |
| Spectrum-Qwen3-14BFeatures=MSP, Protocol=oracle gated2026.06 | 79.6 | |
| Spectrum-Llama-3.1-8BFeatures=MI, Protocol=latent gated2026.06 | 79.5 | |
| Qwen3-14BFeatures=all UQ, Protocol=baseline2026.06 | 79.3 | |
| Spectrum-Qwen3-14BFeatures=MSP, Protocol=baseline2026.06 | 79.3 |