Internal Beakr dataset — not for public distribution. Host as a private repository.
10 multi-hop questions for Pillar 3: knowledge graph / retrieval / citations. The pillar question: does structured retrieval (Beakr's knowledge graph) match or beat flat chunk-based RAG on hard cross-document questions, with traceable citations?
Flavors
control, bridge, comparison, aggregation… See the full description on the dataset page:
https://huggingface.co/datasets/David-beakr/kg-retrieval-seed.