Selective Prediction on TriviaQA (test)
100Power (α=0.1)LEC
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| LECLLM=Qwen2.5-14B2025.12 | 100 | 71.93 | 100 | 100 | 100 | |
| UCB-CLPLLM=LLaMA-3.1-70B2025.12 | 100 | 99.35 | 100 | 100 | 100 | |
| UCB-HFDLLM=LLaMA-3.1-70B2025.12 | 100 | 95.03 | 100 | 100 | 100 | |
| LECLLM=LLaMA-3.1-70B2025.12 | 100 | 99.96 | 100 | 100 | 100 | |
| UCB-CLPLLM=Qwen2.5-14B2025.12 | 99.87 | 62.4 | 100 | 100 | 100 | |
| UCB-HFDLLM=Qwen2.5-14B2025.12 | 97.18 | — | 100 | 100 | 100 | |
| LECLLM=LLaMA-3.1-8B2025.12 | 96.12 | 75.38 | 100 | 100 | 100 | |
| LECLLM=OpenChat-3.52025.12 | 95.21 | 72.3 | 100 | 100 | 100 | |
| UCB-CLPLLM=LLaMA-3.1-8B2025.12 | 93.96 | 71.43 | 100 | 100 | 100 | |
| LECLLM=Vicuna-13B-V1.52025.12 | 93.42 | 65.45 | 100 | 100 | 100 | |
| UCB-CLPLLM=OpenChat-3.52025.12 | 92.94 | 66.84 | 100 | 100 | 100 | |
| UCB-HFDLLM=LLaMA-3.1-8B2025.12 | 90.39 | 53.39 | 100 | 100 | 100 | |
| UCB-CLPLLM=Vicuna-13B-V1.52025.12 | 89.44 | 56.02 | 100 | 100 | 100 | |
| UCB-HFDLLM=OpenChat-3.52025.12 | 88.84 | 60.91 | 100 | 100 | 100 | |
| LECLLM=Qwen2.5-7B2025.12 | 88.5 | 48.89 | 100 | 100 | 100 | |
| UCB-HFDLLM=Vicuna-13B-V1.52025.12 | 83.64 | 52.41 | 100 | 100 | 100 | |
| UCB-CLPLLM=Qwen2.5-7B2025.12 | 83.31 | 39.05 | 100 | 100 | 100 | |
| UCB-HFDLLM=Qwen2.5-7B2025.12 | 73.96 | — | 99.9 | 100 | 100 | |
| LECLLM=Vicuna-7B-V1.52025.12 | 65.08 | — | 92.08 | 100 | 100 | |
| LECLLM=Qwen2.5-3B2025.12 | 59.98 | 27.06 | 90.81 | 100 | 100 | |
| UCB-CLPLLM=Vicuna-7B-V1.52025.12 | 56.86 | — | 86.3 | 100 | 100 | |
| UCB-CLPLLM=Qwen2.5-3B2025.12 | 52.19 | 23.76 | 85.54 | 100 | 100 | |
| UCB-HFDLLM=Vicuna-7B-V1.52025.12 | 42.28 | — | 80.68 | 99.99 | 100 | |
| UCB-HFDLLM=Qwen2.5-3B2025.12 | 38.82 | — | 77.72 | 100 | 100 |