positivethoughts/rewrite_500_prompts_3k_texts

Generated data for Prompt recovery challenge. 3000 examples for now, will share more soon.

500 Prompts created using ChatGPT(4) - Different initial prompts were used to get some variety in creativity, text processing, or professional settings.

To have more variety in the input text I used Wikipedia first paragraphs as the original texts.
https://huggingface.co/datasets/abokbot/wikipedia-first-paragraph

Model used to generate is gemma-7b-it-quant

The prompts file is shared for convinience, along with sharpened cosine similarity scores of each prompt with the baseline "Improve the text"

If you find the dataset useful, please upvote.

https://www.kaggle.com/datasets/dipamc77/3000-rewritten-texts-prompt-recovery-challenge?select=prompts_0_500_wiki_first_para_3000.csv

原始信息汇总

数据集概述

数据集描述

• 数据集名称: Prompt recovery challenge
• 数据量: 目前包含3000个示例，未来将分享更多。

数据生成

• 生成方式: 使用ChatGPT(4)创建了500个不同的初始提示，以获取在创造性、文本处理或专业设置中的多样性。
• 原始文本来源: 使用维基百科的首段作为原始文本。
• 模型: 用于生成的模型是gemma-7b-it-quant。

数据文件

• 文件内容: 包含每个提示与基准“Improve the text”的锐化余弦相似度得分。

数据文件链接

• 文件链接: prompts_0_500_wiki_first_para_3000.csv