Struct QuantizedLinear4BitSpec 

pub struct QuantizedLinear4BitSpec {
    pub in_dim: u32,
    pub out_dim: u32,
    pub weight_type: DataType,
}

Typed metadata for fused grouped INT4 linear.

The actual packed weight buffer is still addressed as u32 words because the kernel extracts eight nibbles per word. This spec binds that physical layout to the first-class DataType::Quantized contract so call sites do not pass an untyped integer buffer and lose the scale/zero-point semantics.

Fields

in_dim: u32

Input feature dimension.

out_dim: u32

Output feature dimension.

weight_type: DataType

First-class quantized weight metadata.

Implementations

impl QuantizedLinear4BitSpec

pub fn affine_grouped(in_dim: u32, out_dim: u32, group_size: u32) -> Self

Build a grouped affine INT4 metadata spec.

Trait Implementations

impl Clone for QuantizedLinear4BitSpec

fn clone(&self) -> QuantizedLinear4BitSpec

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for QuantizedLinear4BitSpec

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

Formats the value using the given formatter.

impl Eq for QuantizedLinear4BitSpec

impl PartialEq for QuantizedLinear4BitSpec

fn eq(&self, other: &QuantizedLinear4BitSpec) -> bool

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for QuantizedLinear4BitSpec