State Stream Transformer (SST) V2: Parallel Training of Nonlinear Recurrence for Latent Space Reasoning
SST V2 introduces a State Stream Transformer that enables continuous latent deliberation by integrating an FFN-driven nonlinear recurrence into every decoder layer. This mechanism allows the model to stream reasoning history horizontally across positions, exploring distinct semantic basins in the latent space. It provides a new, efficient path for complex reasoning beyond traditional scaling methods.