ToolScale is a large-scale synthetic dataset released by NVIDIA in 2025. The associated paper is available at ToolOrchestra: Elevating Intelligence via Efficient Model and Tool Orchestration, aiming to enhance the tool orchestration capabilities of intelligent agents.

The dataset focuses on tool use and reinforcement learning scenarios, enhancing the tool calling and planning capabilities of large models in complex tasks through synthetic data. The dataset contains rich structured fields, covering task descriptions, user scenarios, initial states, and detailed evaluation criteria, supporting tool calling tasks across multiple domains and types.

Dataset Composition

The dataset adopts a structured format and primarily includes the following core fields:

• id: Unique identifier for the data sample.
• description: Includes purpose, relevant_policies, and notes.
• user_scenario: Defines the user scenario, including persona, instructions, and domain. The instructions section is subdivided into task instructions, calling reasons, known information, and unknown information.
• initial_state: Initial state information.
• evaluation_criteria: Evaluation criteria, primarily consisting of an actions list. Each action contains a detailed arguments structure, covering parameters for various tool calls, such as clubs, seasons, games, players, addresses, financial information, travel information, student course selection, payment information, and traffic information. Data types include strings, booleans, integers, floats, and timestamps.

Citation

@misc{toolorchestra,
      title={ToolOrchestra: Elevating Intelligence via Efficient Model and Tool Orchestration}, 
      author={Hongjin Su and Shizhe Diao and Ximing Lu and Mingjie Liu and Jiacheng Xu and Xin Dong and Yonggan Fu and Peter Belcak and Hanrong Ye and Hongxu Yin and Yi Dong and Evelina Bakhturina and Tao Yu and Yejin Choi and Jan Kautz and Pavlo Molchanov},
      year={2025},
      eprint={2511.21689},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2511.21689}, 
}