A neural architecture modeling sequences using continuous state representations, offering efficient processing of long sequences. SSMs like Mamba are alternatives to transformers for applications requiring very long context windows or efficient inference.
See: Context window; Mamba; Transformer