Function Calling on BFCL (SR and Execution Metrics)
77Success Rate (SR)ours
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| oursModel=Qwen3-8B2026.06 | 77 | — | — | — | — | — | |
| GRPOModel=Qwen3-8B2026.06 | 74 | — | — | — | — | — | |
| oursModel=Qwen3-4B2026.06 | 71 | — | — | — | — | — | |
| GRPOModel=Qwen3-4B2026.06 | 69 | — | — | — | — | — | |
| Qwen3-14B + SelSkill Round2Backbone=Qwen3-14B, Training Stage=SelSkill Round22026.05 | 24.2 | 73.5 | 18.2 | 32.4 | 1.01 | 14.2 | |
| Qwen3-14B + SelSkill Round1Backbone=Qwen3-14B, Training Stage=SelSkill Round12026.05 | 22.6 | 69.7 | 14.8 | 30.8 | 0.92 | 14.1 | |
| Qwen3-14B + Base w/ SkillBackbone=Qwen3-14B, Training Stage=Base w/ Skill2026.05 | 18.5 | 44 | 13.4 | 22.8 | 0.73 | 18.2 | |
| Qwen3-14B (No-Skill)Backbone=Qwen3-14B, Training Stage=No-Skill2026.05 | 14.1 | — | — | — | — | 24.2 |