//! Attention mechanisms for time series models.
//!
//! This module provides memory-efficient attention implementations suitable
//! for long time series sequences, including FlashAttention-style tiled
//! online softmax computation that avoids materializing the full N×N matrix.
pub use ;