Adapter fine-tuning: Model customization via standard, bottleneck, or parallel adapters. Hyperformers use task-specific hypernetworks.