Tool Calling on BFCL Multiple v1 (test)
92AccuracyFull-FT
Evaluation results
| Method | Links | |
|---|---|---|
| Full-FTBackbone=Qwen3-8B-Base, KV Sharing=Not Supported2026.02 | 92 | |
| PrefillShareBackbone=Qwen3-8B-Base, KV Sharing=Supported2026.02 | 91 | |
| PrefillShareBackbone=LLaMA3.1-8B, KV Sharing=Supported2026.02 | 88.5 | |
| Full-FTBackbone=LLaMA3.1-8B, KV Sharing=Not Supported2026.02 | 88 | |
| Qwen3-8B-BaseKV Sharing=Inherent2026.02 | 80 | |
| LLaMA3.1-8BKV Sharing=Inherent2026.02 | 45.5 |