DistilGPT-2

Model multiplicity defends small language models against edge device attacks

Researchers have developed a novel defense system called "model multiplicity" to detect adversarial attacks during the training of small language models on edge devices. This approach involves training multiple language…