NesTools: A Dataset for Evaluating Nested Tool Learning Abilities of Large Language Models
This addresses a gap in benchmarking nested tool learning for LLMs, which is incremental as it provides a new dataset for an existing challenge.
The authors tackled the under-explored problem of evaluating large language models' abilities in nested tool learning, where tools are called in sequences with dependencies, by introducing NesTools, a dataset generated through an automatic method and refined manually, and found that current LLMs struggle with this complex task in experiments on 22 models.
Large language models (LLMs) combined with tool learning have gained impressive results in real-world applications. During tool learning, LLMs may call multiple tools in nested orders, where the latter tool call may take the former response as its input parameters. However, current research on the nested tool learning capabilities is still under-explored, since the existing benchmarks lack relevant data instances. To address this problem, we introduce NesTools to bridge the current gap in comprehensive nested tool learning evaluations. NesTools comprises a novel automatic data generation method to construct large-scale nested tool calls with different nesting structures. With manual review and refinement, the dataset is in high quality and closely aligned with real-world scenarios. Therefore, NesTools can serve as a new benchmark to evaluate the nested tool learning abilities of LLMs. We conduct extensive experiments on 22 LLMs, and provide in-depth analyses with NesTools, which shows that current LLMs still suffer from the complex nested tool learning task.