핵심 개념
ToolNet is a plug-and-play framework that organizes massive tools into a directed graph, enabling large language models to efficiently interact with a wide array of tools.
초록
Abstract:
Large language models (LLMs) struggle to effectively utilize massive external tools.
ToolNet proposes a framework that organizes tools into a directed graph for efficient tool selection by LLMs.
Introduction:
Interest in connecting LLMs with tools for real-world tasks is growing.
Existing approaches are limited in connecting LLMs with a small number of tools.
Problem Formulation:
LLMs interact with the environment through structured texts and tools.
ToolNet organizes tools into a graph for LLMs to navigate and select tools efficiently.
ToolNet:
Tools are organized into a directed graph, and LLMs navigate through the graph to select tools.
ToolNet adapts tool transition weights for efficient tool selection.
Experiments:
ToolNet outperforms existing methods in task-specific and multi-task datasets.
ToolNet achieves greater token efficiency and resilience against tool failures.
통계
ToolNet는 대규모 언어 모델이 방대한 도구와 효율적으로 상호 작용할 수 있도록 도구를 유도 그래프로 구성합니다.
ToolNet는 기존 방법보다 작업별 및 다작업 데이터 세트에서 우수한 성능을 발휘합니다.
ToolNet는 더 큰 토큰 효율성과 도구 실패에 대한 저항력을 달성합니다.
인용구
"ToolNet proposes a framework that organizes tools into a directed graph for efficient tool selection by LLMs."
"ToolNet outperforms existing methods in task-specific and multi-task datasets."