Tag
This paper introduces Attractor Models, which use fixed-point solving and implicit differentiation for efficient iterative refinement, achieving superior language modeling and reasoning performance with reduced computational costs compared to traditional transformers.