Saddle Point

Transformers from Spin Models: Approximate Free Energy Minimization

How far can we push the idea of transformers as physical systems?