Natural Language Understanding on MMLU (Accuracy)
80.15Accuracy (MMLU)Base(instruct)
Evaluation Results
| Method | Links | |
|---|---|---|
| Base(instruct)Architecture=Qwen3-30B-A3B2026.05 | 80.15 | |
| GRPOArchitecture=Qwen3-30B-A3B2026.05 | 79.92 | |
| MESAArchitecture=Qwen3-30B-A3B2026.05 | 79.31 | |
| Stair-DPOArchitecture=Qwen3-30B-A3B2026.05 | 79.06 | |
| Stair-SFTArchitecture=Qwen3-30B-A3B2026.05 | 78.94 | |
| SafeXArchitecture=Qwen3-30B-A3B2026.05 | 78.37 | |
| SFTArchitecture=Qwen3-30B-A3B2026.05 | 78.04 | |
| SafeXArchitecture=DeepSeek-v2-Lite2026.05 | 54.25 | |
| GRPOArchitecture=DeepSeek-v2-Lite2026.05 | 54.14 | |
| Base(chat)Architecture=DeepSeek-v2-Lite2026.05 | 54.11 | |
| Stair-DPOArchitecture=DeepSeek-v2-Lite2026.05 | 54.08 | |
| MESAArchitecture=DeepSeek-v2-Lite2026.05 | 53.92 | |
| SFTArchitecture=DeepSeek-v2-Lite2026.05 | 53.71 | |
| Stair-SFTArchitecture=DeepSeek-v2-Lite2026.05 | 53.71 |