Function Calling on BFCL Non-Live v3
94Overall AccuracyQwen2.5-14B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen2.5-14BSpeedup ratio=1×2026.02 | 94 | 99.3 | |
| Qwen3-4BSpeedup ratio=1×2026.02 | 93.7 | 99.3 | |
| ST-Qwen2.5-7BSpeedup ratio=4.5×2026.02 | 93.5 | 99.8 | |
| Qwen2.5-7BSpeedup ratio=1×2026.02 | 93 | 99.7 | |
| Qwen2.5-3BSpeedup ratio=1×2026.02 | 92.5 | 99.7 | |
| ST-Qwen3-4BSpeedup ratio=3.9×2026.02 | 92.5 | 99.5 | |
| ST-Qwen2.5-14BSpeedup ratio=4.4×2026.02 | 92.2 | 99.7 | |
| ST-Qwen2.5-3BSpeedup ratio=3.8×2026.02 | 90.3 | 99.8 | |
| ST-Qwen2.5-1.5BSpeedup ratio=3.5×2026.02 | 88.8 | 99.8 | |
| Qwen2.5-1.5BSpeedup ratio=1×2026.02 | 86.3 | 98.3 | |
| ST-Qwen2.5-0.5BSpeedup ratio=3.1×2026.02 | 79.8 | 99.8 | |
| Qwen2.5-0.5BSpeedup ratio=1×2026.02 | 78.3 | 98.3 |