ToolMaster supervised tool-use trajectories
The ToolMaster supervised fine-tuning release contains 1,444 queries for learning to try tools and execute them from environment feedback. It is stored as a JSON object keyed by query ID.
How do I read the release?
Read the JSON object and iterate over its values. Each record contains query, query_id, tool_doc and instances. Keep the tool descriptions alongside the interactions when comparing how a model responds to an unfamiliar API.
The paper reports selecting 1,500 ToolBench queries during construction. The 1,444 count here describes the public SFT file. Data construction in Section 4
Data files
Authors’ GitHub data directory · The downloads below use a fixed revision.
Related paper
Teaching LLMs to Learn Tool Trialing and Execution through Environment Interaction · Paper on arXiv