Multi-turn Dialogue Evaluation on MT-Bench OOD
8.64R1 ScorePrompt
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PromptModel=Deepseek-V4-Flash, Category=Vanilla2026.06 | 8.64 | 8.06 | |
| PromptModel=Qwen3-32B, Category=Vanilla2026.06 | 8.58 | 8.06 | |
| PromptModel=Qwen3-8B, Category=Vanilla2026.06 | 8.04 | 7.04 | |
| SFTModel=Qwen3-8B, Category=Supervised2026.06 | 7.96 | 7.43 | |
| PromptModel=Qwen3-4B, Category=Vanilla2026.06 | 7.82 | 6.82 | |
| EvoRubricsModel=Qwen3-4B, Category=Ours2026.06 | 7.74 | 7.11 | |
| RuscaRLModel=Qwen3-4B, Category=RL w/ Rubric2026.06 | 7.64 | 6.71 | |
| OnlineRubricsModel=Qwen3-8B, Category=RL w/ Rubric2026.06 | 7.6 | 6.71 | |
| SFTModel=Qwen3-4B, Category=Supervised2026.06 | 7.5 | 6.44 | |
| EvoRubricsModel=Qwen3-8B, Category=Ours2026.06 | 7.48 | 7.05 | |
| GoldenRubricsModel=Qwen3-8B, Category=RL w/ Rubric2026.06 | 7.25 | 6.63 | |
| OnlineRubricsModel=Qwen3-4B, Category=RL w/ Rubric2026.06 | 6.4 | 4.33 | |
| GoldenRubricsModel=Qwen3-4B, Category=RL w/ Rubric2026.06 | 6.04 | 4.01 | |
| RuscaRLModel=Qwen3-8B, Category=RL w/ Rubric2026.06 | 4.11 | 3.4 |