OpenLLMText Dataset

Yutian Chen, Hao Kang, Yiyan Zhai, Liangze Li, Rita Singh, Bhiksha RajZenodo 2023 · DOI · OpenAlex

Abstract

The dataset contains approximately 300k text entries collected from 5 different sources (Human, ChatGPT, PaLM, LLaMA, GPT2-XL). 60k of them are Human-written, randomly selected from the OpenWebText dataset. These entries are collected from the user generated content from Reddit before 2019. 60k of them are the ChatGPT's (gpt3.5-turbo) paragraph-by-paragraph rephrasing for the human written data. 60k of them are the PaLM's (Pathway Language Model, text-bison-001) paragraph-by-paragraph rephrasing for the human written data. 60k of them are the LLaMA-7B's (Large Language Model Meta AI) paragraph-by-pargraph rephrasing for the human written data. 60k of them are the data adapted from the GPT2-output dataset released by the OpenAI (GPT2-XL).
BibTeX
@inproceedings{chen2023openllmtext,
  author = {Yutian Chen and Hao Kang and Yiyan Zhai and Liangze Li and Rita Singh and Bhiksha Raj},
  title = {{OpenLLMText Dataset}},
  booktitle = {Zenodo},
  year = {2023},
  doi = {10.5281/zenodo.8285326},
}

Loading related content...

Figure