Safety evaluation (Accuracy) on Beaver
84.65Accuracy (Beaver Safety)MCPO-DAPO
Evaluation Results
| Method | Links | |
|---|---|---|
| MCPO-DAPOBackbone=Qwen3-8B2026.05 | 84.65 | |
| MCPO-GRPOBackbone=Qwen3-8B2026.05 | 84.26 | |
| MCPO-GRPOBackbone=Qwen3-4B2026.05 | 83.32 | |
| MCPO-DAPOBackbone=Qwen3-4B2026.05 | 82.61 | |
| MT-GRPOBackbone=Qwen3-8B2026.05 | 81.63 | |
| DAPOBackbone=Qwen3-8B2026.05 | 81.15 | |
| MGSBackbone=Qwen3-4B2026.05 | 80.4 | |
| MGSBackbone=Qwen3-8B2026.05 | 80.23 | |
| GRPOBackbone=Qwen3-8B2026.05 | 80.12 | |
| MT-GRPOBackbone=Qwen3-4B2026.05 | 79.93 | |
| CLIPOBackbone=Qwen3-8B2026.05 | 78.86 | |
| GRPOBackbone=Qwen3-4B2026.05 | 78.19 | |
| DAPOBackbone=Qwen3-4B2026.05 | 78.08 | |
| CLIPOBackbone=Qwen3-4B2026.05 | 76.94 | |
| Base ModelBackbone=Qwen3-8B2026.05 | 64.27 | |
| Base ModelBackbone=Qwen3-4B2026.05 | 58.4 |