Once discretized, the state space model is a linear recurrence over the input. Mamba computes this scan efficiently, and the parameters vary per token, which is what makes it selective.
Implement ssm_scan(a, b, x), the recurrence over an input sequence.
Start the state at zero, then for each input update it as a times the state plus b times the input.
No results yet
Run your code to check it against the examples, or submit to run every test.