LME‑MC10 is a 500‑item multiple‑choice benchmark derived from LongMemEval(s).Each item probes one of LongMemEval’s five long‑term memory abilities, but is reformatted into a 10‑option MC task for straightforward automated evaluation (plain accuracy, balanced accuracy, etc.).
Information Extraction (IE)
Multi-Session Reasoning (MR)
Knowledge Updates (KU)
Temporal Reasoning (TR)
Abstention (ABS)
The original AI‑judge rubric is removed;… See the full description on the dataset page:
https://huggingface.co/datasets/Percena/lme-mc10.
