Sweaterdog/openprose

OpenProse is a writing-focused dataset compiled from 5 sources to maximize creative writing skills in Large Language Models.

Datasets used:

• enPurified/smoltalk-creative-writing-enPurified-openai-messages
• Dampfinchen/Creative_Writing_Multiturn
• ChaoticNeutrals/Creative_Writing-ShareGPT
• nothingiisreal/Reddit-Dirty-And-WritingPrompts
• andrewelawrence/writingPrompts-merged

Methodology:

This is a lot of data, and I am well aware. I chose all of these not because I just stumbled upon them, but for one big reason in the dataset by nothingiisreal and andrewelawrence, they are all made by real humans on Reddit.

I chose the name OpenProse since it's targeted towards writing, and a good prose is used in writing, and it's a play off words with OpenPose.

Credit to all of those who made the datasets before me, I just cleaned them and shoved them all into a single, ShareGPT-style dataset for super easy usage.