Medical Question Answering on PubmedQA+HeadQA
52.4AccuracyDUET-IF
Evaluation results
| Method | Links | |
|---|---|---|
| DUET-IFBackbone=Llama-3-8b-Instruct2025.02 | 52.4 | |
| DUET-IFBackbone=Llama-3-8b-Instruct2025.02 | 52.4 | |
| DUET-LESSBackbone=Llama-3-8b-Instruct2025.02 | 50.8 | |
| HeadQATrain=HeadQA, Backbone=Llama-3-8b-Instruct2025.02 | 50.2 | |
| PubmedTrain=Pubmed, Backbone=Llama-3-8b-Instruct2025.02 | 49.3 | |
| AioliBackbone=Llama-3-8b-Instruct2025.02 | 48.8 | |
| TriviaQATrain=TriviaQA, Backbone=Llama-3-8b-Instruct2025.02 | 48.1 | |
| IF onlyBackbone=Llama-3-8b-Instruct, M=200002025.02 | 47.7 | |
| CommonsenseQATrain=CommonsenseQA, Backbone=Llama-3-8b-Instruct2025.02 | 47.2 | |
| Multi-FidBackbone=Llama-3-8b-Instruct2025.02 | 47.2 | |
| ODMBackbone=Llama-3-8b-Instruct2025.02 | 45.8 | |
| SciQTrain=SciQ, Backbone=Llama-3-8b-Instruct2025.02 | 44.6 | |
| Uniform + more training tokensBackbone=Llama-3-8b-Instruct2025.02 | 44.5 | |
| TruthfulQATrain=TruthfulQA, Backbone=Llama-3-8b-Instruct2025.02 | 43.8 | |
| gsm8kTrain=gsm8k, Backbone=Llama-3-8b-Instruct2025.02 | 43.3 | |
| HellaswagTrain=Hellaswag, Backbone=Llama-3-8b-Instruct2025.02 | 43.2 | |
| RandomBackbone=Llama-3-8b-Instruct, M=500002025.02 | 41.2 | |
| WikitextTrain=Wikitext, Backbone=Llama-3-8b-Instruct2025.02 | 40.6 |