💥 Flash Linear Attention brings together hardware-efficient building blocks, training-ready layers, and components for modern sequence models, spanning linear attention, sparse attention, state space ...