Struct ScaleMaskSoftmaxParams 

pub struct ScaleMaskSoftmaxParams {
    pub rows: u32,
    pub cols: u32,
    pub seq_q: u32,
    pub scale: f32,
}

Host-side parameters for scale_mask_softmax_f32.

Fields

rows: u32

Number of rows = n_heads * seq_q (one threadgroup per row).

cols: u32

Length of the reduction axis = seq_k. Must match mask row length.

seq_q: u32

Number of query rows per head (= seq_q). Lets the kernel derive q = row_idx % seq_q for the shared mask index.

scale: f32

Pre-softmax multiplicative scale (e.g. 1.0 / sqrt(head_dim)).

Trait Implementations

impl Clone for ScaleMaskSoftmaxParams

fn clone(&self) -> ScaleMaskSoftmaxParams

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ScaleMaskSoftmaxParams

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

Formats the value using the given formatter.

impl Copy for ScaleMaskSoftmaxParams