Role-Playing on RPGBench In-distribution
-0.034R-EMIRL
Evaluation Results
| Method | Links | |
|---|---|---|
| RLBackbone=Meta-Llama3-8B2025.12 | -0.034 | |
| CoRLBackbone=Qwen3-4B2025.12 | -0.04 | |
| CoRLBackbone=Meta-Llama3-8B2025.12 | -0.049 | |
| RLBackbone=Qwen3-4B2025.12 | -0.05 | |
| SFTBackbone=Meta-Llama3-8B2025.12 | -0.052 | |
| ICLBackbone=Qwen3-4B2025.12 | -0.065 | |
| SFTBackbone=Qwen3-4B2025.12 | -0.067 | |
| DABackbone=Meta-Llama3-8B2025.12 | -0.079 | |
| DABackbone=Qwen3-4B2025.12 | -0.09 | |
| ICLBackbone=Meta-Llama3-8B2025.12 | -0.096 | |
| ThinkingSFTBackbone=Meta-Llama3-8B2025.12 | -0.107 | |
| ThinkingSFTBackbone=Qwen3-4B2025.12 | -0.121 | |
| ThinkingRLBackbone=Qwen3-4B2025.12 | -0.196 | |
| ThinkingRLBackbone=Meta-Llama3-8B2025.12 | -0.272 | |
| DPOBackbone=Qwen3-4B2025.12 | -0.287 | |
| RFTBackbone=Meta-Llama3-8B2025.12 | -0.303 | |
| DPOBackbone=Meta-Llama3-8B2025.12 | -0.586 | |
| RFTBackbone=Qwen3-4B2025.12 | -0.835 |