Multiple Choice Question Answering on MC9, GEN5, AGIEval, BBH, MMLU, MMLU-Pro (test)
69.36MC9 AccuracyFlexMoRE-a4
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| FlexMoRE-a4Params=10.75B, Active Experts=4, Expert Selection Strategy=heterogeneous (MC9)2026.02 | 69.36 | 48.96 | 41.4 | 42.55 | 55.16 | 25.64 | 47.18 | 3.79 | |
| FlexMoRE-a4Params=14.84B, Active Experts=4, Expert Selection Strategy=heterogeneous (all)2026.02 | 68.96 | 48.46 | 41.21 | 41.24 | 55.57 | 25.33 | 46.79 | 2.94 | |
| FlexMoRE-a7Params=10.75B, Active Experts=7, Expert Selection Strategy=heterogeneous (MC9)2026.02 | 68.94 | 48.96 | 41.4 | 42.55 | 55.16 | 25.64 | 47.11 | 14.08 | |
| FlexMoRE-a7Params=14.84B, Active Experts=7, Expert Selection Strategy=heterogeneous (all)2026.02 | 68.62 | 48.34 | 40.42 | 38.61 | 53.98 | 24.2 | 45.69 | 10.65 | |
| FlexMoRE-a2Params=10.75B, Active Experts=2, Expert Selection Strategy=heterogeneous (MC9)2026.02 | 68.38 | 48.68 | 41.84 | 42.69 | 55.55 | 25.43 | 47.1 | 5.49 | |
| FlexMoRE-a4Params=16.21B, Active Experts=4, LoRA Rank (r)=2^10, Expert Selection Strategy=homogeneous2026.02 | 68.19 | 47.84 | 39.2 | 38.87 | 54.63 | 24.87 | 45.6 | 0.32 | |
| FlexMoRE-a2Params=14.84B, Active Experts=2, Expert Selection Strategy=heterogeneous (all)2026.02 | 67.76 | 47.8 | 41.69 | 41.95 | 54.75 | 25.37 | 46.55 | 4.27 | |
| FlexMoRE-a7Params=16.21B, Active Experts=7, LoRA Rank (r)=2^10, Expert Selection Strategy=homogeneous2026.02 | 67.33 | 47.76 | 36.8 | 36.62 | 52.09 | 22.65 | 43.88 | 6.25 | |
| FlexMoRE-a2Params=11.75B, Active Experts=2, LoRA Rank (r)=2^9, Expert Selection Strategy=homogeneous2026.02 | 67.22 | 48.86 | 39.76 | 38.97 | 54.83 | 25.01 | 45.77 | 2.53 | |
| FlexOlmo-a4 BaselineParams=33.27B, Active Experts=4, Expert Selection Strategy=Baseline (full-sized)2026.02 | 67.09 | 45.48 | 39.99 | 40.14 | 55.18 | 24.86 | 45.46 | — | |
| FlexOlmo-a7 BaselineParams=33.27B, Active Experts=7, Expert Selection Strategy=Baseline (full-sized)2026.02 | 65.5 | 36.14 | 37.45 | 35.6 | 50.97 | 22.12 | 41.3 | — | |
| FlexOlmo-a2 (baseline)Params=33.27B, Active Experts=2, Expert Selection Strategy=Baseline (full-sized)2026.02 | 62.57 | 45.08 | 38.42 | 44.69 | 52.97 | 24.15 | 44.65 | — |