Tool Calling on BFCL Non-live
96.67Multiple Success RateParaTool (ours)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ParaTool (ours)Model=Qwen2.5-7B-Instruct2026.05 | 96.67 | 89.83 | 89.5 | 92 | |
| Context+Docs & ExamplesModel=Qwen2.5-7B-Instruct2026.05 | 96.17 | 88.83 | 87.5 | 90.83 | |
| ParaTool (ours)Model=Llama3.1-8B-Instruct2026.05 | 95.33 | 89.33 | 89.5 | 91.39 | |
| Context+DocsModel=Qwen2.5-7B-Instruct2026.05 | 95.33 | 88.17 | 84.17 | 89.22 | |
| Context+Docs & ExamplesModel=Llama3.1-8B-Instruct2026.05 | 95.17 | 88.5 | 87 | 90.22 | |
| Context+DocsModel=Llama3.1-8B-Instruct2026.05 | 95 | 88.33 | 85.5 | 89.61 | |
| Parameter+ToolRetrieverModel=Qwen2.5-7B-Instruct2026.05 | 92.83 | — | 69.33 | 84 | |
| Parameter+Tool RetrieverModel=Llama3.1-8B-Instruct2026.05 | 92.5 | — | 64.33 | 82.06 | |
| Parameter+ReinvokeModel=Qwen2.5-7B-Instruct2026.05 | 92.5 | — | 69.67 | 84 | |
| Parameter+ReinvokeModel=Llama3.1-8B-Instruct2026.05 | 91.83 | — | 66.33 | 82.5 | |
| Parameter+EmbsimModel=Qwen2.5-7B-Instruct2026.05 | 91.83 | — | 69 | 83.56 | |
| Parameter+EmbsimModel=Llama3.1-8B-Instruct2026.05 | 91.33 | — | 63.5 | 81.39 | |
| ToolLLaMAModel=Qwen2.5-7B-Instruct2026.05 | 89.5 | 72.17 | 58.33 | 73.33 | |
| ToolLLaMAModel=Llama3.1-8B-Instruct2026.05 | 87 | 72 | 58.5 | 72.5 | |
| Global ParameterizationModel=Qwen2.5-7B-Instruct2026.05 | 83.83 | 72.83 | 53.33 | 70 | |
| Global ParameterizationModel=Llama3.1-8B-Instruct2026.05 | 62.67 | 72.5 | 46.83 | 60.67 |