Datasets:

LiquidAI
/

phone-assistant-tools

license: other
language:
  - en
  - ar
  - zh
  - fr
  - de
  - it
  - ja
  - ko
  - pt
  - es
  - pl
  - vi
  - th
  - id
  - hi
  - ru
task_categories:
  - text-generation
tags:
  - function-calling
  - tool-use
  - agents
  - on-device
  - grpo
  - reinforcement-learning
size_categories:
  - 1K<n<10K
configs:
  - config_name: default
    data_files:
      - split: train
        path: train.parquet
      - split: test
        path: test.parquet
      - split: test_unseen
        path: test_unseen.parquet
      - split: showcase
        path: showcase.parquet

Phone Assistant Tools

📓 Tutorial: This is the dataset used in GRPO with TRL

Requests to a phone assistant with the gold tool calls, for a fixed catalog of 108 tools. Depending on the request, the right behavior is one or more calls, a clarifying question, or an answer without any tool. 16 languages.

Files

File	Rows	Content
`tools.json`	108	tool catalog, as function schemas
`scoring.json`	108	for each tool argument, how a predicted value is compared with the gold one
`train`	2,500	requests for 98 of the tools
`test`	499	requests for the same 98 tools
`test_unseen`	300	requests for the 10 tools that never appear in `train` or `test` calls
`showcase`	8	examples from `test_unseen`

Fields

Field	Description
`id`	row identifier
`language`	ISO 639-1 code of the request
`category`	`single`, `multi`, `lookalike`, `no_tool`, `clarify` or `partial`
`context`	`{datetime, timezone, locale}` of the phone
`request`	user request
`calls`	gold calls as a JSON string, in any order, or `[]` when no call should be made

Conventions

Dates and times are local and resolved from context.datetime: YYYY-MM-DDTHH:MM:SS, YYYY-MM-DD, HH:MM.
Optional arguments appear only when the user states them.
Argument keys and enum values are English. Free text stays in the user's language.