Synthetic Tool-Use Training Data for Agent Behavioral Traits
Synthetic multi-turn tool-use conversations designed for mechanistic interpretability research on LLM agent behaviors. Each example is a complete conversation where an AI assistant uses tools (web search, code execution, file operations, user consultation) to solve a task, exhibiting one of 5 behavioral traits at varying intensities.
Training pipeline: zactheaipm/qwenscope
