Tool Learning on StableToolBench Average
70.3SoPRGPT-4 (DFSDT)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT-4 (DFSDT)Backbone=GPT-4, Reasoning Strategy=DFSDT2025.01 | 70.3 | 64.2 | |
| GPT-4 (Parallel)Backbone=GPT-4, Reasoning Strategy=Parallel2025.01 | 69.2 | 70.7 | |
| GPT-3.5 (DFSDT)Backbone=GPT-3.5, Reasoning Strategy=DFSDT2025.01 | 66.7 | 65.5 | |
| DTA-LlamaBackbone=Llama-2-7B, Reasoning Strategy=Parallel, Training Protocol=Fine-tuned on DTA-Tool2025.01 | 66.1 | 59.1 | |
| Qwen2.5 (Parallel)Backbone=Qwen2.5-7B-Instruct, Reasoning Strategy=Parallel2025.01 | 63 | 55.3 | |
| GPT-3.5 (Parallel)Backbone=GPT-3.5, Reasoning Strategy=Parallel2025.01 | 61.9 | 53 | |
| ToolLLAMA (DFSDT)Backbone=Llama-2-7B, Reasoning Strategy=DFSDT2025.01 | 54.2 | 47.1 | |
| GPT-4 (ReAct)Backbone=GPT-4, Reasoning Strategy=ReAct2025.01 | 48.2 | 58.7 | |
| GPT-3.5 (ReAct)Backbone=GPT-3.5, Reasoning Strategy=ReAct2025.01 | 47.9 | — | |
| ToolLLaMA† (DFSDT)Backbone=Llama-2-7B, Reasoning Strategy=DFSDT, Training Protocol=DTA-Tool*2025.01 | 39.2 | 37.6 | |
| ToolLLAMA (ReAct)Backbone=Llama-2-7B, Reasoning Strategy=ReAct2025.01 | 37.9 | 39.3 | |
| LLMCompilerBackbone=Llama-2-7B, Reasoning Strategy=Parallel2025.01 | 36.2 | 37.9 | |
| ToolLLaMA† (ReAct)Backbone=Llama-2-7B, Reasoning Strategy=ReAct, Training Protocol=DTA-Tool*2025.01 | 25.3 | 27.3 |