Google Discovered a Technique to Make Native AI As much as 3x Sooner—No New {Hardware} Required
Briefly Google launched Multi-Token Prediction (MTP) drafters for Gemma 4, delivering as much as a 3x speedup at inference with none degradation in output high...