pub fn gated_mla_attend(
q: &mut [f32],
k: &mut [f32],
v: &[f32],
g: &[f32],
t: usize,
cfg: &MlaConfig,
pos0: usize,
theta: f32,
) -> Vec<f32>Expand description
Gated NoPE attend: optionally skip RoPE, SDPA, optional output gate.