ai-model-compression

Researchers from MIT CSAIL and other institutions introduced CompreSSM, a technique that compresses state-space AI models during training by removing unnecessary components early, resulting in faster training and smaller models without sacrificing performance.