DiffusionGemma: 4x faster text generation
Google DeepMind unveils DiffusionGemma, a diffusion-based model generating text 4x faster.
Google DeepMind announced DiffusionGemma, applying diffusion techniques to text generation for a claimed 4x speedup over conventional autoregressive decoding. Coming from a power player, a faster generation paradigm in the open Gemma family is a notable signal for inference efficiency and the viability of diffusion language models.