Tag
This paper introduces SELR, a unified framework for self-explainable latent reasoning that trains a single model to perform efficient reasoning while generating human-readable explanations, eliminating the need for external decoders.