Ethical Reasoning on ETHICS (test)
86AccuracyMultiRole-R1
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MultiRole-R1Backbone=R1-Distill-Qwen-14B, Method Strategy=SelfConsis SFT+GRPO(RS)2025.07 | 86 | 92.89 | |
| SelfConsis SFT+GRPOBackbone=R1-Distill-Qwen-14B2025.07 | 83.5 | 89.42 | |
| MultiRole-R1Backbone=Qwen3-8B, Method Strategy=SelfConsis SFT+GRPO(RS)2025.07 | 81.95 | 89.09 | |
| SelfConsis SFT+DPOBackbone=R1-Distill-Qwen-14B2025.07 | 81.88 | 42.32 | |
| SelfConsis SFTBackbone=R1-Distill-Qwen-14B2025.07 | 81.45 | 71.34 | |
| SelfConsis SFT+GRPOBackbone=Qwen3-8B2025.07 | 79.82 | 86.84 | |
| More thinkBackbone=R1-Distill-Qwen-14B2025.07 | 79.36 | 83.33 | |
| SelfConsis SFTBackbone=Qwen3-8B2025.07 | 78.39 | 68.35 | |
| More thinkBackbone=Qwen3-8B2025.07 | 78.26 | 72.45 | |
| SelfConsis SFT+DPOBackbone=Qwen3-8B2025.07 | 78.09 | 50.31 | |
| Role-PlayBackbone=R1-Distill-Qwen-14B2025.07 | 77.16 | 75.3 | |
| Self-RefineBackbone=R1-Distill-Qwen-14B2025.07 | 76.48 | 83.22 | |
| MultiRole-R1Backbone=R1-Distill-Llama-8B, Method Strategy=SelfConsis SFT+GRPO(RS)2025.07 | 75.84 | 96.54 | |
| SelfConsis SFT+GRPOBackbone=R1-Distill-Llama-8B2025.07 | 75.63 | 87.89 | |
| Zero-shot CoTBackbone=R1-Distill-Qwen-14B2025.07 | 73.63 | 72.2 | |
| Zero-shot CoTBackbone=Qwen3-8B2025.07 | 72.29 | 76.68 | |
| SelfConsis SFT+DPOBackbone=R1-Distill-Llama-8B2025.07 | 71.22 | 83.64 | |
| Role-PlayBackbone=Qwen3-8B2025.07 | 70.67 | 47.83 | |
| Self-RefineBackbone=Qwen3-8B2025.07 | 70.64 | 49.6 | |
| SelfConsis SFTBackbone=R1-Distill-Llama-8B2025.07 | 70.56 | 81.36 | |
| SelfConsis SFT+GRPOBackbone=R1-Distill-Qwen-7B2025.07 | 69.5 | 85.4 | |
| More thinkBackbone=R1-Distill-Llama-8B2025.07 | 68.06 | 83.99 | |
| SelfConsis SFTBackbone=R1-Distill-Qwen-7B2025.07 | 67.45 | 82.19 | |
| SelfConsis SFT+DPOBackbone=R1-Distill-Qwen-7B2025.07 | 67.28 | 68.51 | |
| MultiRole-R1Backbone=R1-Distill-Qwen-7B, Method Strategy=SelfConsis SFT+GRPO(RS)2025.07 | 66.83 | 87.27 | |
| More thinkBackbone=R1-Distill-Qwen-7B2025.07 | 64.44 | 81.53 | |
| Role-PlayBackbone=R1-Distill-Llama-8B2025.07 | 64.24 | 79.78 | |
| Zero-shot CoTBackbone=R1-Distill-Llama-8B2025.07 | 62.46 | 79.44 | |
| Self-RefineBackbone=R1-Distill-Llama-8B2025.07 | 60.96 | 80.17 | |
| Self-RefineBackbone=R1-Distill-Qwen-7B2025.07 | 52.19 | 37.36 | |
| Zero-shot CoTBackbone=R1-Distill-Qwen-7B2025.07 | 51.82 | 36.14 | |
| Role-PlayBackbone=R1-Distill-Qwen-7B2025.07 | 50.83 | 37.89 |