Reasoning on MMLU-Pro (Min/Max/Avg/Voting/ETTC)
94.21Average ScoreQwen-3-Thinking Ensemble
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen-3-Thinking EnsembleModels=30B, 235B2026.05 | 94.21 | 94.12 | 94.31 | 95.04 | 0.9519 | |
| Qwen-3-Thinking EnsembleModels=4B, 30B, 235B2026.05 | 89.86 | 81.16 | 94.31 | 94.82 | 0.9482 | |
| Qwen-3-Thinking EnsembleModels=4B, 235B2026.05 | 87.73 | 81.16 | 94.31 | 89.79 | 0.9467 | |
| Qwen-3-Thinking EnsembleModels=4B, 30B2026.05 | 87.64 | 81.16 | 94.12 | 89.34 | 0.9408 | |
| TeacherBackbone=Llama3.1-70B, Training Method=Supervised Fine-Tuning (SFT), Training Dataset=Full Tulu 32026.06 | 57.37 | — | — | — | — | |
| NeuFSModel=Qwen3-4B, InfoType=Neuron2026.07 | 45.2 | — | — | — | — | |
| TypiClustModel=Qwen3-4B, InfoType=S.2026.07 | 43.7 | — | — | — | — | |
| Highest EntropyModel=Qwen3-4B, InfoType=Highest2026.07 | 43 | — | — | — | — | |
| FastVoteKModel=Qwen3-4B, InfoType=S.+L.2026.07 | 42 | — | — | — | — | |
| NeuFSModel=Qwen3-8B, InfoType=Neuron2026.07 | 41.8 | — | — | — | — | |
| PatronModel=Qwen3-8B, InfoType=S.+E.2026.07 | 41.6 | — | — | — | — | |
| Diverse EntropyModel=Qwen3-4B, InfoType=Diverse2026.07 | 41.4 | — | — | — | — | |
| RandomModel=Qwen3-4B, InfoType=N/A2026.07 | 41.2 | — | — | — | — | |
| FastVoteKModel=Qwen3-8B, InfoType=S.+L.2026.07 | 40.2 | — | — | — | — | |
| VoteKModel=Qwen3-8B, InfoType=S.+E.+L.2026.07 | 40.2 | — | — | — | — | |
| FastVoteKModel=Qwen3-4B, InfoType=L.2026.07 | 40.1 | — | — | — | — | |
| VoteKModel=Qwen3-4B, InfoType=S.+E.2026.07 | 40.1 | — | — | — | — | |
| TypiClustModel=Qwen3-8B, InfoType=S.2026.07 | 39.8 | — | — | — | — | |
| TypiClustModel=Qwen3-4B, InfoType=S.+L.2026.07 | 39.8 | — | — | — | — | |
| Highest EntropyModel=Qwen3-8B, InfoType=Highest2026.07 | 39.4 | — | — | — | — | |
| Diverse EntropyModel=Qwen3-8B, InfoType=Diverse2026.07 | 39.2 | — | — | — | — | |
| VoteKModel=Qwen3-8B, InfoType=S.+E.2026.07 | 39.1 | — | — | — | — | |
| PatronModel=Qwen3-4B, InfoType=S.+E.2026.07 | 39.1 | — | — | — | — | |
| RandomModel=Qwen3-8B, InfoType=N/A2026.07 | 38.8 | — | — | — | — | |
| FastVoteKModel=Qwen3-4B, InfoType=S.2026.07 | 38.6 | — | — | — | — | |
| TypiClustModel=Qwen3-8B, InfoType=S.+L.2026.07 | 38.3 | — | — | — | — | |
| TypiClustModel=Qwen3-8B, InfoType=L.2026.07 | 38.2 | — | — | — | — | |
| FastVoteKModel=Qwen3-8B, InfoType=S.2026.07 | 37.9 | — | — | — | — | |
| TypiClustModel=Qwen3-4B, InfoType=L.2026.07 | 37.6 | — | — | — | — | |
| VoteKModel=Qwen3-4B, InfoType=S.+E.+L.2026.07 | 37.4 | — | — | — | — | |
| FastVoteKModel=Qwen3-8B, InfoType=L.2026.07 | 37.3 | — | — | — | — | |
| GKDBackbone=Llama3.1-8B, Training Method=Two-stage KD strategy (GKD), Training Dataset=Full Tulu 32026.06 | 35.16 | — | — | — | — | |
| SFTBackbone=Llama3.1-8B, Training Method=Supervised Fine-Tuning (SFT), Training Dataset=Full Tulu 32026.06 | 34.65 | — | — | — | — | |
| SeqKDBackbone=Llama3.1-8B, Training Method=Two-stage KD strategy (SeqKD), Training Dataset=Full Tulu 32026.06 | 34.64 | — | — | — | — | |
| NeuFSModel=Llama-3.1-8B, InfoType=Neuron2026.07 | 32.7 | — | — | — | — | |
| RandomModel=Llama-3.1-8B, InfoType=N/A2026.07 | 32.5 | — | — | — | — | |
| Diverse EntropyModel=Llama-3.1-8B, InfoType=Diverse2026.07 | 32.4 | — | — | — | — | |
| TypiClustModel=Llama-3.1-8B, InfoType=S.2026.07 | 32.3 | — | — | — | — | |
| FastVoteKModel=Llama-3.1-8B, InfoType=L.2026.07 | 32.3 | — | — | — | — | |
| TypiClustModel=Llama-3.1-8B, InfoType=L.2026.07 | 32.2 | — | — | — | — | |
| TypiClustModel=Llama-3.1-8B, InfoType=S.+L.2026.07 | 31.8 | — | — | — | — | |
| FastVoteKModel=Llama-3.1-8B, InfoType=S.+L.2026.07 | 31.8 | — | — | — | — | |
| VoteKModel=Llama-3.1-8B, InfoType=S.+E.2026.07 | 31.8 | — | — | — | — | |
| Highest EntropyModel=Llama-3.1-8B, InfoType=Highest2026.07 | 31.7 | — | — | — | — | |
| FastVoteKModel=Llama-3.1-8B, InfoType=S.2026.07 | 31.6 | — | — | — | — | |
| PatronModel=Llama-3.1-8B, InfoType=S.+E.2026.07 | 31.3 | — | — | — | — | |
| VoteKModel=Llama-3.1-8B, InfoType=S.+E.+L.2026.07 | 30.9 | — | — | — | — | |
| NeuFSModel=Llama-3.2-3B, InfoType=Neuron2026.07 | 25.1 | — | — | — | — | |
| RandomModel=Llama-3.2-3B, InfoType=N/A2026.07 | 24.9 | — | — | — | — | |
| TypiClustModel=Llama-3.2-3B, InfoType=L.2026.07 | 24.8 | — | — | — | — | |
| VoteKModel=Llama-3.2-3B, InfoType=S.+E.2026.07 | 24.6 | — | — | — | — | |
| Highest EntropyModel=Llama-3.2-3B, InfoType=Highest2026.07 | 24.5 | — | — | — | — | |
| FastVoteKModel=Llama-3.2-3B, InfoType=S.2026.07 | 24.5 | — | — | — | — | |
| PatronModel=Llama-3.2-3B, InfoType=S.+E.2026.07 | 24.4 | — | — | — | — | |
| Diverse EntropyModel=Llama-3.2-3B, InfoType=Diverse2026.07 | 24.4 | — | — | — | — | |
| TypiClustModel=Llama-3.2-3B, InfoType=S.+L.2026.07 | 24.3 | — | — | — | — | |
| TypiClustModel=Llama-3.2-3B, InfoType=S.2026.07 | 24.2 | — | — | — | — | |
| FastVoteKModel=Llama-3.2-3B, InfoType=L.2026.07 | 24.2 | — | — | — | — | |
| FastVoteKModel=Llama-3.2-3B, InfoType=S.+L.2026.07 | 24.2 | — | — | — | — | |
| VoteKModel=Llama-3.2-3B, InfoType=S.+E.+L.2026.07 | 23.3 | — | — | — | — |