Necent/

efficientrag-filter-training-data

Training data for the Filter component of EfficientRAG.

JSONL with fields:

query_info — concatenation of original question + extracted useful tokens token_labels — per-word binary labels (1=keep, 0=discard)

	Statistics




	

Count

Total samples 5,691

	Data Sources




	

Source Language Samples Method

HotpotQA (5K questions) EN ~5K Heuristic labels

Dragon-derec multi-hop (690) RU ~700… See the full description on the dataset page: https://huggingface.co/datasets/Necent/efficientrag-filter-training-data.