ResearchBenchmarksSequence Sorting on Toy Vocabulary Sort 1K sequences (held-out)Follow99.15AccuracyBaseline98.765298.865198.96599.0649Apr 18, 2026Evaluation ResultsMethodMethodLinksAccuracyPerplexityActive Fraction (α)TLOps SavedBaselineModel Dimension (d)=12...Model Dimension (d)=128, Number of Layers (L)=62026.0499.151.079100—TSAModel Dimension (d)=12...Model Dimension (d)=128, Number of Layers (L)=62026.0498.781.0877322.5