Home

Awesome

ToolEyes

ToolEyes: Fine-Grained Evaluation for Tool Learning Capabilities of Large Language Models in Real-world Scenarios

Data for paper ToolEyes: Fine-Grained Evaluation for Tool Learning Capabilities of Large Language Models in Real-world Scenarios

Junjie Ye

jjye23@m.fudan.edu.cn

Jan. 01, 2024

Introduction

Existing evaluations of tool learning primarily focus on validating the alignment of selected tools for large language models (LLMs) with expected outcomes. However, these approaches rely on a limited set of scenarios where answers can be pre-determined, diverging from genuine needs. Furthermore, a sole emphasis on outcomes disregards the intricate capabilities essential for LLMs to effectively utilize tools. To tackle this issue, we propose ToolEyes, a fine-grained system tailored for the evaluation of the LLMs' tool learning capabilities in authentic scenarios. The system meticulously examines seven real-world scenarios, analyzing five dimensions crucial to LLMs in tool learning: format alignment, intent comprehension, behavior planning, tool selection, and answer organization. Additionally, ToolEyes incorporates a tool library boasting approximately 600 tools, serving as an intermediary between LLMs and the physical world. Evaluations involving ten LLMs across three categories reveal a preference for specific scenarios and limited cognitive abilities in tool learning. Intriguingly, expanding the model size even exacerbates the hindrance to tool learning. These findings offer instructive insights aimed at advancing the field of tool learning.

<div> <center> <img src=Figures/ToolEyes.png> </div>

What's New

Requirement

Data

The tool lobrary and test data are released, which can be found in ToolEyes/Tool_Library and /ToolEyes/Test_Data, respectively. Below is the statistics of the data:

ScenarioTGDURSPLIRAMFTTotal
# Cat556896241
# Subcat6514301971495
# Tool27267516415016496568
# Query58495670544550382

Results in Different Scenarios

We evaluate the tool learning performance of the LLMs across seven real-world scenarios.

<div> <center> <img src=Figures/result-scenario.png> </div>

Results of Different LLMs Capabilities

We examine the entirety of the tool learning process, focusing on the five dimensions of capability essential for LLMs to successfully undertake tool learning.

<div> <center> <img src=Figures/result-capability.png> </div>

Citation

If you find this project useful in your research, please cite:

@article{DBLP:journals/corr/abs-2401-00741,
  author       = {Junjie Ye and
                  Guanyu Li and
                  Songyang Gao and
                  Caishuang Huang and
                  Yilong Wu and
                  Sixian Li and
                  Xiaoran Fan and
                  Shihan Dou and
                  Qi Zhang and
                  Tao Gui and
                  Xuanjing Huang},
  title        = {ToolEyes: Fine-Grained Evaluation for Tool Learning Capabilities of
                  Large Language Models in Real-world Scenarios},
  journal      = {CoRR},
  volume       = {abs/2401.00741},
  year         = {2024},
  url          = {https://doi.org/10.48550/arXiv.2401.00741},
  doi          = {10.48550/ARXIV.2401.00741},
  eprinttype    = {arXiv},
  eprint       = {2401.00741},
  timestamp    = {Mon, 05 Feb 2024 20:18:16 +0100},
  biburl       = {https://dblp.org/rec/journals/corr/abs-2401-00741.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}