Gemma 4 met multi-token voorspelling genereert tokens tot drie keer sneller

Google's Gemma 4 model incorporates multi-token prediction technology that accelerates inference speed by generating several tokens simultaneously through speculative decoding. The approach enables the model to validate multiple predictions in a single computational pass, achieving roughly threefold faster processing without compromising output quality.