Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Toolken+: Improving LLM Tool Usage with Reranking and a Reject Option

Konstantin Yakovlev, Sergey Nikolenko, Andrey Bout

TL;DR

This work introduces Toolken+ that mitigates the first problem by reranking top $k$ tools selected by ToolkenGPT and the second problem with a special"Reject"option such that the model will generate a vocabulary token if"Reject"is ranked first.

Abstract

The recently proposed ToolkenGPT tool learning paradigm demonstrates promising performance but suffers from two major issues: first, it cannot benefit from tool documentation, and second, it often makes mistakes in whether to use a tool at all. We introduce Toolken+ that mitigates the first problem by reranking top $k$ tools selected by ToolkenGPT and the second problem with a special "Reject" option such that the model will generate a vocabulary token if "Reject" is ranked first. We demonstrate the effectiveness of Toolken+ on multistep numerical reasoning and tool selection tasks.

Toolken+: Improving LLM Tool Usage with Reranking and a Reject Option

TL;DR

This work introduces Toolken+ that mitigates the first problem by reranking top tools selected by ToolkenGPT and the second problem with a special"Reject"option such that the model will generate a vocabulary token if"Reject"is ranked first.

Abstract

The recently proposed ToolkenGPT tool learning paradigm demonstrates promising performance but suffers from two major issues: first, it cannot benefit from tool documentation, and second, it often makes mistakes in whether to use a tool at all. We introduce Toolken+ that mitigates the first problem by reranking top tools selected by ToolkenGPT and the second problem with a special "Reject" option such that the model will generate a vocabulary token if "Reject" is ranked first. We demonstrate the effectiveness of Toolken+ on multistep numerical reasoning and tool selection tasks.

Paper Structure

This paper contains 7 sections, 1 theorem, 6 equations, 1 figure, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Experiments
  5. Conclusion
  6. Limitations
  7. Appendix: Prompts for Toolken+

Key Result

Proposition 1

The following is a computationally stable upper bound of eq:opt1, up to an additive constant independent of $\mathbf{W}_{\mathcal{T}'}$:

Figures (1)

  • Figure 1: Toolken+ sample operation.

Theorems & Definitions (2)

  • Proposition 1: Naive upper bound
  • proof