pub fn sdpa( q: &[f32], k: &[f32], v: &[f32], t: usize, heads: usize, dq: usize, dv: usize, ) -> Vec<f32>
Causal scaled-dot-product attention. q/k: [T, H, dq], v: [T, H, dv].
q/k
[T, H, dq]
v
[T, H, dv]