Entropica: 1024-mode unitary evolution with Born-rule readout, trained on TinyStories in under 2 hours on a laptop GPU

Entropica is the first generative language model whose forward pass is physically realizable as a passive linear-optical interferometer (zero electrical power during inference).  
A 1024-mode, 32-layer unitary network using only Reck-scheme MZI meshes and Born-rule readout learns coherent TinyStories-style generation in under 1.8 hours on a single laptop GPU.  
Model is ready for a full optical implementation path demonstrated with printed phase masks and a $30 laser pointer.  
All code, weights, and dataset generation scripts are public.