How Gemma 4’s multi-token prediction and community-driven DFlash are speeding up local LLM throughput by 3-6x. The post Google brings multi-token prediction Gemma 4 LLMs first appeared on TechTalks.