Question Answering on TriviaQA (EM, F1, Retrieval Rate, Delta Latency)
62.2EMMargin
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MarginBackbone=Qwen2.5-7B-Instruct, tau=0.152025.11 | 62.2 | 62.6 | 33.8 | 2.174 | |
| EntropyBackbone=Qwen2.5-7B-Instruct, tau=0.802025.11 | 61.8 | 62.2 | 2.8 | 0.876 | |
| VarianceBackbone=Qwen2.5-7B-Instruct, tau=0.752025.11 | 61.8 | 62.2 | 0.6 | 0.133 | |
| NeverBackbone=Qwen2.5-7B-Instruct2025.11 | 60.8 | 61.4 | 0 | 2.947 | |
| AlwaysBackbone=Qwen2.5-7B-Instruct2025.11 | 57.6 | 57.2 | 100 | 3.462 |