saxenan3/

temporalwiki-drift-cl

Fact-update continual-learning benchmark built from monthly Wikipedia snapshots (Nov 2025 → Feb 2026). For each adjacent snapshot pair we extract the drift set of (subject, relation, object) triples whose object value changed, plus a held-out stable set of facts that did not change across the chain (a forgetting probe).

	Splits




	

split rows description

train 1500 drift triples to learn, 500 per slice (s1: Nov→Dec… See the full description on the dataset page: https://huggingface.co/datasets/saxenan3/temporalwiki-drift-cl.