pub fn launch_k3_mla_decode_token(
gpu: &dyn GpuBackend,
kernels: &K3MlaDecodeKernels,
q: &mut [f32],
k: &mut [f32],
v: &[f32],
g: &[f32],
kv: &mut MlaKv,
cfg: &MlaConfig,
pos: usize,
theta: f32,
stream: u64,
) -> Result<Vec<f32>>Expand description
2026-09-25: One decode token against a host MlaKv: writes the roped
q/k back, appends to kv, and uploads the whole KV history.