The dataset is derived from BOOKSUM
The idea here is to make use of the BOOKSUM dataset to finetune models with larger context length (8k+) but very few samples in BOOKSUM have such length.
Enter multi-chapter summaries!
The context here comprises multiple chapters taken from the same book appended together to form a larger context length.
The prompt requests a summary from one of the chapters and a summary of the corresponding chapter is present in the… See the full description on the dataset page:
https://huggingface.co/datasets/shahules786/Multi-chapter-summaries.
