Dataset Card for dataforge-economics

Table of Contents

• Overview
• Dataset Description
  • Data Collection and Synthesis
  • Data Structure
• Licensing, Privacy, and Ethics
• Access
• Usage
• Citation
• Contributions

Overview

This dataset, teknium/dataforge-economics, is a specialized collection of 1,000 synthetic examples in the field of economics. It has been generated using OpenAI's GPT-4 and a custom data synthesis pipeline named DataForge, developed by me.

Dataset Description

Data Collection and Synthesis

The data in teknium/dataforge-economics has been synthetically generated using OpenAI's GPT-4 language model. The synthesis process was enhanced and structured using the DataForge pipeline, which incorporates domain-specific knowledge and ensures relevance in economics topics.

Data Structure

• Size of dataset: 1000 examples
• Type of data: Textual (Economics domain-specific)
• Data format: JSON
• Fields:
• • id: a randomly generated uuid
  • conversations: single turn human & gpt turns in sharegpt format
  • source: the dataset name itself, for metadata purposes when merging with others
  • topic: the sub-topic for the domain
  • system_prompt: type of system prompt used for generating the response.

Licensing, Privacy, and Ethics

• License: MIT License
• Special Considerations: This datasest is purely generated from GPT-4 data, some information may be incorrect or invalid.
• Privacy: As the dataset is synthetically generated, it does not contain any real individual's data.

Access

• Availability: General Access

Usage

This dataset is a domain specialist dataset, the first to use my new pipeline called Data Forge, which can create domain expert knowledge (and tasks, as seen in the Trismegistus occult dataset) This dataset was a proof of concept to improve upon Orca model's economics expertise, which surpassed my custom benchmark for economics when finetuned over stable beluga.