As reported by 9to5mac, Apple has teamed up with The Ohio State University to launch a novel diffusion language model, namely the FS-DFM (Few-Step Discrete Flow Matching Model). Leveraging a three-step training approach, this model makes a significant leap forward in the efficiency of long-text generation. It can produce entire paragraphs with merely 8 iterations, marking a speed that is 128 times greater than that of conventional models.