CoSER is a high-quality dataset for role-playing LLMs, sourced from 771 renowned novels. The dataset contains authentic multi-turn, multi-character dialogues extracted from acclaimed literary works.
Authentic Content: Unlike synthetic datasets, CoSER extracts real dialogues from literature, maintaining high fidelity to the original works. The dialogues are inherently multi-turn and multi-character, exhibiting natural… See the full description on the dataset page:
https://huggingface.co/datasets/Rohan183/CoSER.
