Saved in:
Bibliographic Details
Main Authors: Jang, Kyochul, Lee, Donghyeon, Kim, Kyusik, Heo, Dongseok, Lee, Taewhoo, Kim, Woojeong, Suh, Bongwon
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2506.22853
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915368290746368
author Jang, Kyochul
Lee, Donghyeon
Kim, Kyusik
Heo, Dongseok
Lee, Taewhoo
Kim, Woojeong
Suh, Bongwon
author_facet Jang, Kyochul
Lee, Donghyeon
Kim, Kyusik
Heo, Dongseok
Lee, Taewhoo
Kim, Woojeong
Suh, Bongwon
contents Existing function-calling benchmarks focus on single-turn interactions. However, they overlook the complexity of real-world scenarios. To quantify how existing benchmarks address practical applications, we introduce DICE-SCORE, a metric that evaluates the dispersion of tool-related information such as function name and parameter values throughout the dialogue. Analyzing existing benchmarks through DICE-SCORE reveals notably low scores, highlighting the need for more realistic scenarios. To address this gap, we present DICE-BENCH, a framework that constructs practical function-calling datasets by synthesizing conversations through a tool graph that maintains dependencies across rounds and a multi-agent system with distinct personas to enhance dialogue naturalness. The final dataset comprises 1,607 high-DICE-SCORE instances. Our experiments on 19 LLMs with DICE-BENCH show that significant advances are still required before such models can be deployed effectively in real-world settings. Our code and data are all publicly available: https://snuhcc.github.io/DICE-Bench/.
format Preprint
id arxiv_https___arxiv_org_abs_2506_22853
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DICE-BENCH: Evaluating the Tool-Use Capabilities of Large Language Models in Multi-Round, Multi-Party Dialogues
Jang, Kyochul
Lee, Donghyeon
Kim, Kyusik
Heo, Dongseok
Lee, Taewhoo
Kim, Woojeong
Suh, Bongwon
Computation and Language
Artificial Intelligence
Existing function-calling benchmarks focus on single-turn interactions. However, they overlook the complexity of real-world scenarios. To quantify how existing benchmarks address practical applications, we introduce DICE-SCORE, a metric that evaluates the dispersion of tool-related information such as function name and parameter values throughout the dialogue. Analyzing existing benchmarks through DICE-SCORE reveals notably low scores, highlighting the need for more realistic scenarios. To address this gap, we present DICE-BENCH, a framework that constructs practical function-calling datasets by synthesizing conversations through a tool graph that maintains dependencies across rounds and a multi-agent system with distinct personas to enhance dialogue naturalness. The final dataset comprises 1,607 high-DICE-SCORE instances. Our experiments on 19 LLMs with DICE-BENCH show that significant advances are still required before such models can be deployed effectively in real-world settings. Our code and data are all publicly available: https://snuhcc.github.io/DICE-Bench/.
title DICE-BENCH: Evaluating the Tool-Use Capabilities of Large Language Models in Multi-Round, Multi-Party Dialogues
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2506.22853