mquake-benchmark-two-variants

IN premise — summaries/2026/08/24/zhong-2023-mquake-s5-mello-a-proposal-for-editing-large.md

Created 2026-08-25T02:59:06+00:00

MQuAKE benchmark consists of two subsets: MQuAKE-CF (counterfactual edits) and MQuAKE-T (temporal/real knowledge changes), evaluating multi-hop questions whose answers are entailed consequences of edits.

Summary

The MQuAKE benchmark divides its testing into two flavors: hypothetical "what if" edits to facts, and real-world changes that happened over time. In both cases, the point is to check whether a model can trace how one change ripples through multiple connected pieces of knowledge, rather than just updating a single isolated fact.