Software Engineering on SWE tasks Robustness Fix n = 55 (test)
57.9F1 ScoreParametricSkills
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ParametricSkillsBackbone=Qwen3-8B, Skill Type=Parametric (Hypernetwork generated)2026.06 | 57.9 | 90.17 | 75.09 | |
| In-ContextBackbone=Qwen3-8B, Skill Type=In-context skills2026.06 | 51.13 | 88.83 | 66 | |
| No-SkillBackbone=Qwen3-8B, Skill Type=None2026.06 | 50.79 | 89.37 | 66.91 | |
| SHINEBackbone=Qwen3-8B, Skill Type=Parametric (SHINE checkpoint)2026.06 | 41.26 | 88.56 | 61 |