This is Grammarly's coedit dataset parsed into Alpaca-style instruction, input, and output rows, with the original instruction values replaced with a more diverse set of procedurally generated instructions. Contains 23930 unique values of instruction, as compared to the original 144. See coedit_reword.py for how these were generated.
All credit to the original authors of this dataset.
@article{raheja2023coedit,
title={CoEdIT: Text… See the full description on the dataset page:
https://huggingface.co/datasets/chargoddard/coedit-reworded.
