Instruction Following on IFEval MultiIF
90.65I-Following ScoreA-SFT → A-GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| A-SFT → A-GRPOBackbone=Qwen3-4B2025.09 | 90.65 | |
| SFT → A-GRPOBackbone=Qwen3-4B2025.09 | 90.45 | |
| SFT → GRPOBackbone=Qwen3-4B2025.09 | 89.75 | |
| A-CHORDBackbone=Qwen3-4B2025.09 | 88.27 | |
| A-GRPOBackbone=Qwen3-4B2025.09 | 88.17 | |
| CHORDBackbone=Qwen3-4B2025.09 | 86.97 | |
| GRPOBackbone=Qwen3-4B2025.09 | 86.9 | |
| A-SFTBackbone=Qwen3-4B2025.09 | 82.47 | |
| SFTBackbone=Qwen3-4B2025.09 | 80.6 | |
| Qwen3-4BBackbone=Qwen3-4B2025.09 | 80.1 | |
| A-SFT → A-GRPOBackbone=Qwen3-0.6B2025.09 | 79.86 | |
| A-GRPOBackbone=Qwen3-0.6B2025.09 | 79.58 | |
| SFT → A-GRPOBackbone=Qwen3-0.6B2025.09 | 79.57 | |
| SFT → GRPOBackbone=Qwen3-0.6B2025.09 | 77.4 | |
| A-CHORDBackbone=Qwen3-0.6B2025.09 | 75.25 | |
| GRPOBackbone=Qwen3-0.6B2025.09 | 74.09 | |
| CHORDBackbone=Qwen3-0.6B2025.09 | 73.52 | |
| A-SFTBackbone=Qwen3-0.6B2025.09 | 60.72 | |
| SFTBackbone=Qwen3-0.6B2025.09 | 59.97 | |
| Qwen3-0.6BBackbone=Qwen3-0.6B2025.09 | 56.62 |