Multiple-choice Question Answering on MMLU-Pro (test)
89.6AccuracyEntangle-based Reweight
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Entangle-based ReweightVerifier aggregation approach=Entangle-based Reweight2026.04 | 89.6 | 93.2 | 90.6 | 0.045 | |
| Accuracy-based ReweightVerifier aggregation approach=Accuracy-based Reweight2026.04 | 88.1 | 91.9 | 89.1 | 0.034 | |
| Majority VoteVerifier aggregation approach=Majority Vote2026.04 | 84.7 | 90.1 | — | — | |
| AMOOptimizer=AMO, Model=Llama3.1-1.4B, Few-shot=52026.05 | 7.99 | — | — | — | |
| AMOOptimizer=AMO, Model=Llama3.1-760M, Few-shot=52026.05 | 6.76 | — | — | — |