Google's DiffusionGemma AI model offers 4x speed

Google DeepMind has unveiled DiffusionGemma, an experimental open AI model that significantly accelerates text generation. Unlike traditional autoregressive models that produce text token by token, DiffusionGemma processes entire blocks of tokens in parallel, similar to image diffusion models. This parallel processing allows it to achieve up to four times the speed of existing Gemma models on local hardware, including gaming GPUs and enterprise AI accelerators. The model, a 26 billion parameter Mixture of Experts, is optimized for non-linear tasks like editing and problem-solving, making it a promising development for efficient local AI applications.