NUMB1234/AskToAct-7B

Model Overview

AskToAct-7B is a tool-augmented LLM fine-tuned on Qwen2.5-7B-Instruct, designed to handle real-world scenarios where user queries are often ambiguous or unspecified. It is developed based on our paper: AskToAct: Enhancing LLMs Tool Use via Self-Correcting Clarification.

This model is capable of:

• Detecting unspecified user intent which is essential for tool invocation
• Proactively eliciting user intent through multi-turn clarification
• Recovering from common clarification errors to ensure efficient interaction and accurate final tool invocation

AskToAct-7B represents a step forward in tool-augmented LLMs by systematically incorporating intent clarification into the tool-use process. This integration enhances both the model’s ability to understand unspecified queries and its effectiveness in executing accurate tool calls, resulting in a more natural and efficient user interaction experience.

Citation

If you find this work useful in your method, you can cite the paper as below:

@misc{zhang2025asktoactenhancingllmstool,
      title={AskToAct: Enhancing LLMs Tool Use via Self-Correcting Clarification}, 
      author={Xuan Zhang and Yongliang Shen and Zhe Zheng and Linjuan Wu and Wenqi Zhang and Yuchen Yan and Qiuying Peng and Jun Wang and Weiming Lu},
      year={2025},
      eprint={2503.01940},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2503.01940}, 
}