ResearchBenchmarksZero-shot Commonsense Reasoning Suite (PIQA, HellaSwag, WinoGrande, ARC-E/C, SIQA, BoolQ, LAMBADA)Follow73.3PIQA Accuracy (Zero-shot)OSDN-APF72.67672.8387373.162May 13, 2026Evaluation ResultsMethodMethodLinksPIQA Accuracy (Zero-shot)HellaSwag Accuracy (Zero-shot)WinoGrande Accuracy (Zero-shot)ARC-Easy Accuracy (Zero-shot)ARC-Challenge Accuracy (Zero-shot)SIQA Accuracy (Zero-shot)BoolQ Accuracy (Zero-shot)LAMBADA Accuracy (Zero-shot)Average Accuracy (Zero-shot)OSDN-APFModel Size=1.3B, Train...Model Size=1.3B, Training Tokens=100B, zero-shot=true2026.0573.359.260.66940.441.760.54856.6DeltaNetModel Size=1.3B, Train...Model Size=1.3B, Training Tokens=100B, zero-shot=true2026.0572.758.159.868.741.841.557.947.356