Gemma 4 MLX setup hits 19 tok/s

Strong niche utility for people who want a no-frills local LLM setup on Apple Silicon, but it reads more like a practical benchmark note than a polished product launch.

• –The main signal is performance: ~19 tok/s on an M4 MacBook with 16GB is credible and useful for local-model shoppers.
• –The setup choice matters: Flask + plain HTML lowers complexity and makes the workflow easier to reproduce than a full desktop app stack.
• –Passing full conversation history each turn is good for narrative work, but it will pressure memory and context efficiency as chats grow.
• –The 4-bit question is the real open item; long-context behavior is where these lightweight local setups usually start to trade quality for speed.