Probabilistic Tiny Recursive Model

Probabilistic TRM (PTRM) extends Tiny Recursive Models with stochastic exploration to escape suboptimal fixed points during iterative latent-state refinement. Unlike prior TRM workarounds requiring task-specific input perturbations, PTRM is task-agnostic and targets test-time compute scaling. Relevant to practitioners seeking parameter-efficient reasoning models with more reliable convergence behavior.