←── back to feed
/topics/diffusiongemma-26b-moe-model-launch

DiffusionGemma 26B MoE model launch

4 items1 sourcesupdated 6d agotrend 0

Google DeepMind released DiffusionGemma, a 26B mixture-of-experts open model that applies diffusion techniques to text generation, achieving up to 4x faster output by generating multiple words in parallel rather than sequentially. NVIDIA has optimized the model to run on GeForce RTX, RTX PRO, and DGX Spark systems.

  • 26B mixture-of-experts open model from Google DeepMind
  • Generates multiple words in parallel instead of one-at-a-time, enabling 4x faster text generation
  • Optimized by NVIDIA for GeForce RTX, RTX PRO, and DGX Spark systems
  • Applies diffusion techniques—traditionally used in image generation—to text output for low-latency inference