Enum AttentionStrategy 

pub enum AttentionStrategy {
    FlashBlackboxAccelerated(BlackboxAcceleratedStrategy),
    FlashUnit,
    Fallback,
    Autotune,
}

Strategy used to select which attention implementation to run.

Variants

FlashBlackboxAccelerated(BlackboxAcceleratedStrategy)

Flash Attention using accelerated inner matmuls.

FlashUnit

Flash Attention using unit inner matmuls.

Fallback

Fallback implementation using multiple separate kernels.

Autotune

Available on crate feature autotune only.

Automatically benchmark and select the best strategy at runtime.

Trait Implementations

impl Debug for AttentionStrategy

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for AttentionStrategy

fn default() -> Self

Returns the “default value” for a type.