This dataset contains scored and assigned nuggets for 5k single-turn battles based on the lmarena-ai/search-arena-v1-7k dataset.
For each battle, gpt-4.1 is used to generate nuggets of information from the two model responses. Each nugget is scored based on its importance, and then the same model is used to assign the nuggets to the two model responses.
In addition to the original fields from lmarena-ai/search-arena-v1-7k, each data point includes:
nugget_documents: A list… See the full description on the dataset page:
https://huggingface.co/datasets/castorini/search-arena-v1-nuggets-5k.
