Enum SparsityPattern 

pub enum SparsityPattern {
    Unstructured,
    NM {
        n: usize,
        m: usize,
    },
    Block {
        height: usize,
        width: usize,
    },
    Row,
    Column,
}

Sparsity pattern constraints.

Defines the structural constraints on which weights can be pruned. Different patterns offer different trade-offs between flexibility and hardware acceleration.

Variants

Unstructured

No structural constraint - any element can be pruned.

Maximum flexibility but requires sparse hardware for speedup.

NM

N:M sparsity - in every M consecutive elements, exactly N are non-zero.

Hardware Support

• 2:4 sparsity: NVIDIA Ampere (A100, RTX 30xx) - 2x speedup
• 4:8 sparsity: Future hardware

Fields

n: usize

Number of non-zero elements per group

m: usize

Group size

Block

Block sparsity - entire blocks of size (height, width) are pruned together.

Useful for structured pruning where entire neurons or filters are removed.

Fields

height: usize

Block height

width: usize

Block width

Row

Row sparsity - entire rows (output channels) pruned.

Equivalent to pruning entire output neurons.

Column

Column sparsity - entire columns (input channels) pruned.

Equivalent to removing input features.

Implementations

impl SparsityPattern

pub fn is_valid(&self) -> bool

Check if this pattern configuration is valid.

Returns

true if the pattern parameters are valid.

pub fn theoretical_sparsity(&self) -> Option<f32>

Get the theoretical sparsity for this pattern.

Returns

Sparsity ratio (0.0 = dense, 1.0 = fully sparse)

pub fn validate(&self, mask: &Tensor) -> Result<(), PruningError>

Validate a mask tensor against this pattern’s constraints.

Arguments

• mask - Binary mask tensor to validate

Returns

Ok(()) if valid, Err(PruningError::InvalidPattern) if not.

Trait Implementations

impl Clone for SparsityPattern

fn clone(&self) -> SparsityPattern

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SparsityPattern

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

Formats the value using the given formatter.

impl Default for SparsityPattern

fn default() -> Self

Returns the “default value” for a type.

impl PartialEq for SparsityPattern

fn eq(&self, other: &SparsityPattern) -> 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 SparsityPattern

impl Eq for SparsityPattern

impl StructuralPartialEq for SparsityPattern