ResearchBenchmarksPhysical Commonsense Reasoning on PIQA (Relative Improvement Metrics)Follow1.89Average Relative ImprovementTBDF-0.63720.01890.6751.3311Jan 29, 2026Evaluation ResultsMethodMethodLinksAverage Relative ImprovementInferior/Superior CountsTBDFFiltering Mode=General...Filtering Mode=General, Models=Gemma 3 (270M, 1B, 4B)2026.011.89—TBDFFiltering Mode=FW-EDU,...Filtering Mode=FW-EDU, Models=Gemma 3 (270M, 1B, 4B)2026.011.35—CBFiltering Mode=FW-EDU,...Filtering Mode=FW-EDU, Models=Gemma 3 (270M, 1B, 4B)2026.01-0.54—