ResearchBenchmarksCommonsense sentence completion on HellaSwag full (test val)Follow26.7AccuracyGPT2-201M24.93225.39125.8526.309May 20, 2026Evaluation ResultsMethodMethodLinksAccuracyGPT2-201Mmode=Zero-shot, parame...mode=Zero-shot, parameters=201M2026.0526.7SNN-194Mmode=Zero-shot, parame...mode=Zero-shot, parameters=194M2026.0526.5Randommode=Zero-shotmode=Zero-shot2026.0525