In-Context Edit

A novel approach that achieves state-of-the-art instruction-based editing using just 0.5% of the training data and 1% of the parameters required by prior SOTA methods.