Tool-calling on Stable Toolbench I2-Cat
78.3Pass RateParaTool
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ParaToolBackbone=Qwen2.5-7B-Instruct2026.05 | 78.3 | 80.19 | |
| ParaToolBackbone=Llama3.1-8B-Instruct2026.05 | 72.64 | 74.53 | |
| Parameter+ReinvokeBackbone=Qwen2.5-7B-Instruct2026.05 | 72.64 | 71.7 | |
| Context+Docs & ExamplesBackbone=Llama3.1-8B-Instruct2026.05 | 69.81 | 67.92 | |
| ToolLLaMABackbone=Qwen2.5-7B-Instruct2026.05 | 68.54 | 72.58 | |
| ToolLLaMABackbone=Llama3.1-8B-Instruct2026.05 | 66.94 | 69.35 | |
| Parameter+ReinvokeBackbone=Llama3.1-8B-Instruct2026.05 | 65.09 | 66.98 | |
| Context+Docs & ExamplesBackbone=Qwen2.5-7B-Instruct2026.05 | 64.15 | 68.87 | |
| Parameter+ToolRetrieverBackbone=Qwen2.5-7B-Instruct2026.05 | 62.62 | 60.38 | |
| Context+DocsBackbone=Llama3.1-8B-Instruct2026.05 | 61.32 | — | |
| Parameter+ToolRetrieverBackbone=Llama3.1-8B-Instruct2026.05 | 61.32 | 62.26 | |
| Parameter+EmbsimBackbone=Qwen2.5-7B-Instruct2026.05 | 58.49 | 61.32 | |
| Parameter+EmbsimBackbone=Llama3.1-8B-Instruct2026.05 | 57.55 | 60.38 | |
| Context+DocsBackbone=Qwen2.5-7B-Instruct2026.05 | 54.72 | — | |
| Global ParameterizationBackbone=Qwen2.5-7B-Instruct2026.05 | 42.45 | 45.28 | |
| Global ParameterizationBackbone=Llama3.1-8B-Instruct2026.05 | 36.79 | 32.08 |