Question Answering on TriviaQA (test) (EM and F1)
35.6EMGround Truth
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ground Truthsetting=SFT settings2025.12 | 35.6 | 44.66 | |
| EPD-Judge Modelsetting=SFT settings2025.12 | 33.2 | 40.47 | |
| DP-Logitssetting=SFT settings2025.12 | 26.9 | 31.5 | |
| LLM-Hampsetting=SFT settings2025.12 | 20.7 | 28.74 |