---
pretty_name: "Tool finetuning dataset"
task_categories:
  - text-generation
language:
  - en
tags:
  - fine-tuning-data
  - synthetic-data
  - json
  - function-calling
  - tool-use
  - llama-3.2-3b
  - together-ai
size_categories:
  - <100K
license: other # Source text license depends on original textbooks. Generated content depends on Llama 3 license and source text.
---

# Tool Finetuning Dataset

## Dataset Description

### Dataset Summary

This dataset is designed for fine-tuning language models to use tools (function calling) appropriately based on user queries. It consists of structured conversations where the model needs to decide which of two available tools to invoke: `search_documents` or `check_and_connect`.

The dataset combines:
1. Adapted natural questions that should trigger the `search_documents` tool
2. System status queries that should trigger the `check_and_connect` tool

### Dataset Structure

The dataset is provided in JSONL format with each entry containing a complete conversation with system, user, and assistant messages.

#### Data Instances

Each data instance follows this structure:
```json
{
  "conversations": [
    {
      "role": "system",
      "content": "Environment: ipython\nCutting Knowledge Date: December 2023\nToday Date: 18 May 2025"
    },
    {
      "role": "user",
      "content": "Given the following functions, please respond with a JSON for a function call with its proper arguments that best answers the given prompt. Respond in the format {\"name\": function name, \"parameters\": dictionary of argument name and its value}. Do not use variables.\n{ \"type\": \"function\", \"function\": { \"name\": \"check_and_connect\", \"description\": \"check_and_connect\", \"parameters\": { \"properties\": {}, \"type\": \"object\" } } }\n{ \"type\": \"function\", \"function\": { \"name\": \"search_documents\", \"description\": \"\\n Searches for documents based on a user's query string. Use this to find information on a specific topic.\\n\\n \", \"parameters\": { \"properties\": { \"query\": { \"description\": \"The actual search phrase or question. For example, 'What are the causes of climate change?' or 'population of Madre de Dios'.\", \"type\": \"string\" } }, \"required\": [ \"query\" ], \"type\": \"object\" } } }\nSearch information about where will you find earth's magnetic north pole"
    },
    {
      "role": "assistant",
      "content": "   {\"name\": \"search_documents\", \"parameters\": {\"query\": \"where will you find earth's magnetic north pole\"}}"
    }
  ]
}

```

For `search_documents` examples:
- User queries are prefixed with "Search information about" 
- Assistant responses call the `search_documents` function with the original query

For `check_and_connect` examples:
- User queries are variants of system status checks
- Assistant responses call the `check_and_connect` function with empty parameters

#### Data Fields

- `conversations`: Array of message objects
  - `role`: String representing the speaker (system, user, or assistant)
  - `content`: String containing the message content

### Dataset Creation

#### Source Data

The dataset is generated from:
1. `maximedb/natural_questions` dataset from Hugging Face (for `search_documents` examples)
2. Predefined list of system status queries (for `check_and_connect` examples)

#### Processing

For `search_documents`:
- 1,000 questions are selected from Natural Questions
- Each question is prefixed with "Search information about"
- The JSON response includes the original question without the prefix

For `check_and_connect`:
- 50 samples are generated from a set of 15 predefined system status queries
- The JSON response has empty parameters: `{"name": "check_and_connect", "parameters": {}}`

The final dataset is shuffled to randomize the order of examples.

### Considerations for Using the Data

#### Discussion of Biases

The dataset may reflect biases inherent in:
- The Natural Questions dataset
- The manual selection of system status queries

#### Dataset Metadata

- **Size**: 1,050 examples (1,000 search_documents + 50 check_and_connect)
- **Format**: JSONL
- **Creation Date**: Generated on May 18, 2025
- **License**: Inherits from Natural Questions dataset
- **Tools**: 
  - `search_documents`: For information retrieval queries
  - `check_and_connect`: For system status checks

## Additional Information

### Citation

If using this dataset, please cite both this work and the original Natural Questions dataset:

```
@misc{tool_finetuning_dataset,
  title={Tool Finetuning Dataset},
  year={2025}
}

@article{47761,
  title = {Natural Questions: a Benchmark for Question Answering Research},
  author = {Tom Kwiatkowski and Jennimaria Palomaki and Olivia Redfield and Michael Collins and Ankur Parikh and Chris Alberti and Danielle Epstein and Illia Polosukhin and Matthew Kelcey and Jacob Devlin and Kenton Lee and Kristina N. Toutanova and Llion Jones and Ming-Wei Chang and Andrew Dai and Jakob Uszkoreit and Quoc Le and Slav Petrov},
  year = {2019},
  journal = {Transactions of the Association of Computational Linguistics}
}
```