Dataset for 'LLM prompt recovery competition). Fields: 'original text', 'rewrite_prompts', 'rewritten_text','id'. I took this data for creating this dataset:

• llmpr-public-10k-unique
• gpt3-condenced-prompt-style
• around 2.5k generated data using gemma, based on @aatiffraz notebook

I just concated all data, dropped dublicates by 'rewrite_prompt'. Also I processed data by deleting "Sure, here's" or "Sure, here is" from beginning with simple regex like r"Sure, .*:" and then r"Here, .*:", removed \n , _ and * from start of message.