Question Answering on TriviaQA Commonsense
69.6AccuracyDUET-IF
Evaluation results
| Method | Links | |
|---|---|---|
| DUET-IFBackbone=Llama-3-8b-Instruct2025.02 | 69.6 | |
| DUET-LESSBackbone=Llama-3-8b-Instruct2025.02 | 67.6 | |
| Multi-FidBackbone=Llama-3-8b-Instruct2025.02 | 65.2 | |
| AioliBackbone=Llama-3-8b-Instruct2025.02 | 63.7 | |
| ODMBackbone=Llama-3-8b-Instruct2025.02 | 60.1 | |
| Uniform + more training tokensBackbone=Llama-3-8b-Instruct2025.02 | 59.2 | |
| IF onlyBackbone=Llama-3-8b-Instruct, M=200002025.02 | 57.8 | |
| RandomBackbone=Llama-3-8b-Instruct, M=500002025.02 | 57.3 |