Multilingual Understanding on CMMLU
81.5ScoreNo train
Evaluation results
| Method | Links | |
|---|---|---|
| No trainBackbone=S1-14B2026.04 | 81.5 | |
| SafeChainBackbone=R1-32B2026.04 | 80.6 | |
| STARBackbone=R1-32B2026.04 | 80.6 | |
| R1-ALTBackbone=R1-32B2026.04 | 80 | |
| No trainBackbone=R1-14B2026.04 | 79.7 | |
| No trainBackbone=R1-32B2026.04 | 79.7 | |
| SafeChainBackbone=R1-14B2026.04 | 78.5 | |
| R1-ALTBackbone=R1-14B2026.04 | 77.3 | |
| SafeChainBackbone=S1-14B2026.04 | 77.3 | |
| STARBackbone=R1-14B2026.04 | 76.1 | |
| R1-ALTBackbone=S1-14B2026.04 | 74 | |
| KEELArchitecture=512 Layers / 3B Params, Peak Learning Rate=4.5 x 10^-3, Pre-training Tokens=1T, Evaluation Protocol (Shots)=5-shot2026.01 | 72 | |
| Qwen2.5-Dense2MoELLM=Qwen2.5, Model=Ours, Activated Parameters=1.2 B2026.05 | 70.5 | |
| Qwen2.5-DenseLLM=Qwen2.5, Model=Dense, Activated Parameters=1.5 B2026.05 | 67.82 | |
| Pre-LNArchitecture=512 Layers / 3B Params, Peak Learning Rate=3.0 x 10^-3, Pre-training Tokens=1T, Evaluation Protocol (Shots)=5-shot2026.01 | 66.6 | |
| No trainBackbone=S1-3B2026.04 | 64.8 | |
| R1-ALTBackbone=S1-3B2026.04 | 62.1 | |
| SafeChainBackbone=S1-3B2026.04 | 56.4 | |
| SafeChainBackbone=R1-7B2026.04 | 53.1 | |
| R1-ALTBackbone=R1-7B2026.04 | 53.1 | |
| No trainBackbone=R1-7B2026.04 | 52.2 | |
| STARBackbone=R1-7B2026.04 | 51.9 | |
| R1-ALTBackbone=R1-8B2026.04 | 47.2 | |
| SafeChainBackbone=R1-8B2026.04 | 45.4 | |
| No trainBackbone=R1-8B2026.04 | 45.1 | |
| STARBackbone=R1-8B2026.04 | 44.2 | |
| Llama2-Dense2MoELLM=Llama2, Model=Ours, Activated Parameters=5.7 B2026.05 | 39.56 | |
| Llama2-DenseLLM=Llama2, Model=Dense, Activated Parameters=7.0 B2026.05 | 32.75 | |
| R1-ALTBackbone=R1-1.5B2026.04 | 30.1 | |
| No trainBackbone=R1-1.5B2026.04 | 29.9 | |
| STARBackbone=R1-1.5B2026.04 | 27.2 | |
| SafeChainBackbone=R1-1.5B2026.04 | 26.9 |