speculative decoding

tech

Google's Gemma 4 AI Models Get 3x Speed Boost by Predicting Future Tokens

Google introduces Multi-Token Prediction for Gemma 4, speeding up local AI generation by up to 3x with speculative decoding.