Last released Oct 15, 2026
A Production-Grade Framework for Subquadratic Attention, Model Compression, and Hardware-Aware Transformer Optimization