Tag
SupraLabs released SupraSafety-18M, a tiny 18M-parameter BERT-style content moderation model trained on NVIDIA's Nemotron-3.5 dataset. It achieves 81.2% accuracy and runs efficiently on edge devices.