Skip to content

Dataset · LiquidAI

Phone assistant tool calling dataset

LiquidAI/phone-assistant-tools

Smartphone assistant prompts in 16 languages with reference calls from a catalog of 108 functions and evaluation rules.

What's inside

Created by LiquidAI for training models using GRPO, this dataset contains between 1,000 and 10,000 user prompts in 16 languages for a smartphone assistant. Depending on the request, the reference action involves calling one or more functions from a fixed catalog (108 tools in tools.json), a clarifying question, or a standard response without an API call. A scoring.json file is also included with the logic for evaluating predicted arguments.

What you can do with it

  • Training LLMs and function-calling agents using reinforcement learning (GRPO) to manage smartphone features
  • Training models to recognize cases when a clarifying question is needed instead of a tool call
  • Automating the evaluation of function selection accuracy and argument filling on multilingual prompts
Tasks
Text generation
Languages
English, Arabic, Chinese, French, German, Italian, Japanese, Korean and 4 more
Size
1K–10K rows
Downloads this month
660
Likes
10

How to load it

from datasets import load_dataset

ds = load_dataset("LiquidAI/phone-assistant-tools")