Struct QuantizePerGroupDescriptor 

pub struct QuantizePerGroupDescriptor {
    pub outer_size: i32,
    pub axis_size: i32,
    pub group_size: i32,
    pub q_min: i32,
    pub q_max: i32,
    pub input_element: ElementKind,
    pub output_element: ElementKind,
}

Descriptor for a quantize_per_group forward op.

Fields

outer_size: i32

Product of all dims except the quant axis (the flattened non-quant prefix).

axis_size: i32

Length of the quant axis. Must be >= 0 and divisible by group_size.

group_size: i32

Group size — number of consecutive elements along the quant axis that share a (scale, zp) pair. Typical: 128 for GPTQ INT4 weights.

q_min: i32

Quantization range lower bound.

q_max: i32

Quantization range upper bound.

input_element: ElementKind

Input FP element kind.

output_element: ElementKind

Output int element kind.

Implementations

impl QuantizePerGroupDescriptor

pub fn num_groups(&self) -> i32

Number of groups along the quant axis. Equals axis_size / group_size (validated axis_size % group_size == 0).

Trait Implementations

impl Clone for QuantizePerGroupDescriptor

fn clone(&self) -> QuantizePerGroupDescriptor

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for QuantizePerGroupDescriptor

impl Debug for QuantizePerGroupDescriptor

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

Formats the value using the given formatter.