Repulsor: Speeding Up Generative Models with Memory

The Repulsor paper introduces a novel contrastive memory bank to accelerate generative modeling. The approach likely offers significant performance improvements by efficiently storing and retrieving relevant information during generation.