Enum SparsePattern 

#[non_exhaustive]
pub enum SparsePattern {
    LocalWindow,
    GlobalLocal,
    Random,
    BlockSparse,
    Sliding,
}

Sparse-attention pattern variants.

Controls which query–key pairs are evaluated during attention. All patterns give sub-quadratic complexity for long sequences.

use scirs2_neural::attention::sparse::SparsePattern;
let p = SparsePattern::LocalWindow;
assert_eq!(p, SparsePattern::LocalWindow);

Variants (Non-exhaustive)

Non-exhaustive enums could have additional variants added in future. Therefore, when matching against variants of non-exhaustive enums, an extra wildcard arm must be added to account for any future variants.

LocalWindow

Sliding-window local attention: each token attends to the window_size tokens immediately before and after it (Longformer local attention).

GlobalLocal

Global + local attention: a set of designated global tokens attend to all positions and all positions attend to global tokens; non-global tokens additionally use a local window (Longformer full model).

Random

Uniform-random attention pattern: each token attends to n_random randomly chosen positions in addition to its local window. Used in the BigBird architecture.

BlockSparse

Block-sparse attention (BigBird): the sequence is divided into blocks of size block_size and each block attends to: itself, n_random_blocks randomly chosen blocks, and any designated global tokens.

Sliding

Alias for LocalWindow — a sliding window with no global tokens.

Trait Implementations

impl Clone for SparsePattern

fn clone(&self) -> SparsePattern

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for SparsePattern

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

Formats the value using the given formatter.

impl Display for SparsePattern

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

Formats the value using the given formatter.

impl PartialEq for SparsePattern

fn eq(&self, other: &SparsePattern) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Copy for SparsePattern

impl Eq for SparsePattern

impl StructuralPartialEq for SparsePattern