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.