Google made Gemma 4 models 3x faster with MTP Drafters

Google introduced speculative decoding for Gemma 4, cutting inference latency by pairing a main model with a lightweight drafter. It speeds local and on-device AI on GPUs and edge hardware without sacrificing output quality.

🗞 #gemini @testingcatalog