Role-Playing on RPGBench Aggregate (Overall)
-0.026Avg ScoreCoRL
Evaluation Results
| Method | Links | |
|---|---|---|
| CoRLBackbone=Qwen3-4B2025.12 | -0.026 | |
| RLBackbone=Qwen3-4B2025.12 | -0.041 | |
| SFTBackbone=Qwen3-4B2025.12 | -0.057 | |
| CoRLBackbone=Meta-Llama3-8B2025.12 | -0.058 | |
| DABackbone=Qwen3-4B2025.12 | -0.062 | |
| RLBackbone=Meta-Llama3-8B2025.12 | -0.087 | |
| DABackbone=Meta-Llama3-8B2025.12 | -0.091 | |
| SFTBackbone=Meta-Llama3-8B2025.12 | -0.093 | |
| ThinkingSFTBackbone=Meta-Llama3-8B2025.12 | -0.107 | |
| ThinkingSFTBackbone=Qwen3-4B2025.12 | -0.113 | |
| ThinkingRLBackbone=Qwen3-4B2025.12 | -0.183 | |
| ThinkingRLBackbone=Meta-Llama3-8B2025.12 | -0.218 | |
| RFTBackbone=Meta-Llama3-8B2025.12 | -0.233 | |
| DPOBackbone=Qwen3-4B2025.12 | -0.253 | |
| ICLBackbone=Qwen3-4B2025.12 | -0.265 | |
| ICLBackbone=Meta-Llama3-8B2025.12 | -0.435 | |
| DPOBackbone=Meta-Llama3-8B2025.12 | -0.489 | |
| RFTBackbone=Qwen3-4B2025.12 | -0.722 |