Tool-use instruction following on ToolBench Average
71.1Pass RateDFSDT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DFSDTBackbone=GPT-4, Retriever=Oracle2023.07 | 71.1 | 70.4 | |
| DFSDTBackbone=ToolLLaMA, Retriever=API Retriever2023.07 | 67.3 | 63.1 | |
| DFSDTBackbone=ToolLLaMA, Retriever=Oracle2023.07 | 66.7 | 60 | |
| DFSDTBackbone=ChatGPT, Retriever=Oracle2023.07 | 64.8 | 64.3 | |
| ReACTBackbone=GPT-4, Retriever=Oracle2023.07 | 57.2 | 64.4 | |
| DFSDTBackbone=Text-Davinci-003, Retriever=Oracle2023.07 | 43.1 | 46.3 | |
| ReACTBackbone=ChatGPT, Retriever=Oracle2023.07 | 40.2 | — | |
| ReACTBackbone=ToolLLaMA, Retriever=Oracle2023.07 | 29 | 47 | |
| DFSDTBackbone=Claude-2, Retriever=Oracle2023.07 | 22.6 | 43.5 | |
| ReACTBackbone=Text-Davinci-003, Retriever=Oracle2023.07 | 16.5 | 33.2 | |
| ReACTBackbone=Claude-2, Retriever=Oracle2023.07 | 6.8 | 34.4 | |
| ReACT & DFSDTBackbone=Vicuna, Retriever=Oracle2023.07 | 0 | 0 | |
| ReACT & DFSDTBackbone=Alpaca, Retriever=Oracle2023.07 | 0 | 0 |