Instruction Following on Instruction Following Suite IFB, IFE
27.3IFB ScoreCARE-RL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CARE-RLBackbone=Qwen3-4B2026.05 | 27.3 | 48.9 | |
| CARE-RLBackbone=Qwen2.5-7B2026.05 | 27.2 | 41.8 | |
| MOPDBackbone=Qwen3-4B2026.05 | 26.9 | 47.6 | |
| NV→VBackbone=Qwen3-4B2026.05 | 26.8 | 46.6 | |
| MGSBackbone=Qwen3-4B2026.05 | 26.3 | 46.9 | |
| MOPDBackbone=Qwen2.5-7B2026.05 | 26.1 | 41.2 | |
| Naive MixingBackbone=Qwen3-4B2026.05 | 26 | 45.7 | |
| V→NVBackbone=Qwen3-4B2026.05 | 25.6 | 45 | |
| MGSBackbone=Qwen2.5-7B2026.05 | 25.3 | 40.4 | |
| V→NVBackbone=Qwen2.5-7B2026.05 | 24.8 | 38.2 | |
| NV→VBackbone=Qwen2.5-7B2026.05 | 24.4 | 39.2 | |
| Naive MixingBackbone=Qwen2.5-7B2026.05 | 24.2 | 38.9 | |
| BaseBackbone=Qwen3-4B2026.05 | 18 | 39.4 | |
| BaseBackbone=Qwen2.5-7B2026.05 | 15 | 31.2 |