Selective Prediction on TriviaQA
0.3234E-AURCSequence NLL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Sequence NLLBase Model=Gemma 3 4B2026.03 | 0.3234 | 0.2807 | |
| Sequence NLLBase Model=Qwen3 4B2026.03 | 0.2594 | 0.2671 | |
| Sequence NLLBase Model=Llama 3.2 3B2026.03 | 0.1944 | 0.1818 | |
| Semantic EntropyBase Model=Gemma 3 4B2026.03 | 0.1703 | 0.2151 | |
| PC ProbeBase Model=Gemma 3 4B2026.03 | 0.1613 | 0.0743 | |
| Combined: PC + NLLBase Model=Gemma 3 4B2026.03 | 0.1583 | 0.0724 | |
| SE ProbeBase Model=Qwen3 4B2026.03 | 0.1413 | 0.0824 | |
| Combined: PC + SEBase Model=Gemma 3 4B2026.03 | 0.1383 | 0.0571 | |
| SE ProbeBase Model=Gemma 3 4B2026.03 | 0.1232 | 0.1413 | |
| SE ProbeBase Model=Llama 3.2 3B2026.03 | 0.1192 | 0.0734 | |
| Combined: PC + SE ProbeBase Model=Gemma 3 4B2026.03 | 0.1192 | 0.0521 | |
| SE ProbeBase Model=Ministral 8B2026.03 | 0.1072 | 0.038 | |
| PC ProbeBase Model=Ministral 8B2026.03 | 0.0992 | 0.035 | |
| Combined: PC + NLLBase Model=Ministral 8B2026.03 | 0.0962 | 0.0351 | |
| Semantic EntropyBase Model=Qwen3 4B2026.03 | 0.0942 | 0.045 | |
| Sequence NLLBase Model=Ministral 8B2026.03 | 0.0912 | 0.031 | |
| PC ProbeBase Model=Llama 3.2 3B2026.03 | 0.0872 | 0.0734 | |
| Combined: PC + NLLBase Model=Llama 3.2 3B2026.03 | 0.0772 | 0.0804 | |
| Combined: PC + SEBase Model=Ministral 8B2026.03 | 0.0732 | 0.029 | |
| Combined: PC + SE ProbeBase Model=Llama 3.2 3B2026.03 | 0.0701 | 0.1031 | |
| Combined: PC + SE ProbeBase Model=Ministral 8B2026.03 | 0.0691 | 0.032 | |
| PC ProbeBase Model=Qwen3 4B2026.03 | 0.0661 | 0.0632 | |
| Semantic EntropyBase Model=Ministral 8B2026.03 | 0.0661 | 0.036 | |
| Combined: PC + NLLBase Model=Qwen3 4B2026.03 | 0.0651 | 0.0622 | |
| Combined: PC + SE ProbeBase Model=Qwen3 4B2026.03 | 0.0571 | 0.0642 | |
| Combined: PC + SEBase Model=Llama 3.2 3B2026.03 | 0.0551 | 0.0491 | |
| Combined: PC + SEBase Model=Qwen3 4B2026.03 | 0.0461 | 0.0773 | |
| Semantic EntropyBase Model=Llama 3.2 3B2026.03 | 0.0451 | 0.0411 |