Struct FlashDecodingArgs 

pub struct FlashDecodingArgs<'a, T: Element> {
    pub q: TensorRef<'a, T, 3>,
    pub k: TensorRef<'a, T, 4>,
    pub v: TensorRef<'a, T, 4>,
    pub y: TensorMut<'a, T, 3>,
}

Args bundle for a FlashDecoding launch.

Q is rank-3 because seq_q == 1 is encoded in the descriptor — no need to thread a unit axis through the API.

K/V take shape [B, H_kv, K_len, D] (the PHYSICAL layout, not the broadcast-replicated H_q view). The kernel handles the Q→KV head mapping via integer division kv_head = q_head / group_size. For pure MHA the caller just passes H_kv == H_q and the same data shape as before.

Fields

q: TensorRef<'a, T, 3>

Query tensor — shape [B, H_q, D]. Arbitrary strides via the supplied stride array; typical case is contig.

k: TensorRef<'a, T, 4>

Key tensor — shape [B, H_kv, K_len, D], physical layout.

v: TensorRef<'a, T, 4>

Value tensor — shape [B, H_kv, K_len, D], physical layout.

y: TensorMut<'a, T, 3>

Output tensor — shape [B, H_q, D].