Chat and Dialogue on Chat Performance Suite WB, CW3, RQA
44.1WB ScoreCARE-RL
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CARE-RLBackbone=Qwen3-4B2026.05 | 44.1 | 49.2 | 47.5 | 46.9 | |
| MOPDBackbone=Qwen3-4B2026.05 | 43.6 | 48.3 | 46.7 | 46.2 | |
| MGSBackbone=Qwen3-4B2026.05 | 42.8 | 46.5 | 45.8 | 45 | |
| V→NVBackbone=Qwen3-4B2026.05 | 42 | 48.1 | 44.5 | 44.9 | |
| Naive MixingBackbone=Qwen3-4B2026.05 | 41.1 | 43.6 | 45.3 | 43.3 | |
| NV→VBackbone=Qwen3-4B2026.05 | 40.5 | 39.8 | 46 | 42.1 | |
| CARE-RLBackbone=Qwen2.5-7B2026.05 | 39.2 | 48.2 | 44.7 | 44 | |
| MOPDBackbone=Qwen2.5-7B2026.05 | 38.5 | 47.1 | 43.8 | 43.1 | |
| MGSBackbone=Qwen2.5-7B2026.05 | 37.4 | 44.5 | 43 | 41.6 | |
| V→NVBackbone=Qwen2.5-7B2026.05 | 36.5 | 46.8 | 40.1 | 41.1 | |
| Naive MixingBackbone=Qwen2.5-7B2026.05 | 35.9 | 39.2 | 42.3 | 39.1 | |
| NV→VBackbone=Qwen2.5-7B2026.05 | 35.7 | 35.4 | 43.9 | 38.3 | |
| BaseBackbone=Qwen3-4B2026.05 | 25.3 | 45 | 43.4 | 37.9 | |
| BaseBackbone=Qwen2.5-7B2026.05 | 22.1 | 43.5 | 41.1 | 35.6 |