ToolSandbox: A Stateful, Conversational, Interactive Evaluation Benchmark for LLM Tool Use Capabilities
This provides a comprehensive evaluation framework for tool-use LLMs, addressing a bottleneck in assessing real-world applications, though it is incremental as it builds on prior work.
The authors tackled the problem of evaluating large language models' tool-use capabilities by introducing ToolSandbox, a benchmark that includes stateful tool execution and conversational evaluation, showing a significant performance gap between open-source and proprietary models and identifying challenging tasks for state-of-the-art models.
Recent large language models (LLMs) advancements sparked a growing research interest in tool assisted LLMs solving real-world challenges, which calls for comprehensive evaluation of tool-use capabilities. While previous works focused on either evaluating over stateless web services (RESTful API), based on a single turn user prompt, or an off-policy dialog trajectory, ToolSandbox includes stateful tool execution, implicit state dependencies between tools, a built-in user simulator supporting on-policy conversational evaluation and a dynamic evaluation strategy for intermediate and final milestones over an arbitrary trajectory. We show that open source and proprietary models have a significant performance gap, and complex tasks like State Dependency, Canonicalization and Insufficient Information defined in ToolSandbox are challenging even the most capable SOTA LLMs, providing brand-new insights into tool-use LLM capabilities. ToolSandbox evaluation framework is released at https://github.com/apple/ToolSandbox