Recurrent Mode (Inference)
During token-by-token generation, apply the recurrence directly: x_t = Āx_{t-1} + B̄u_t.
During token-by-token generation, apply the recurrence directly: x_t = Āx_{t-1} + B̄u_t. Each step is O(1) time and memory (just one matrix-vector multiply). No need to see the full sequence; state is a fixed-size summary.