New AI Models Run Entirely in Your Browser, No Internet Needed
This could mean faster, private AI on your laptop without cloud fees.
Deep Dive
A Hugging Face collection called Bonsai 2 Small is available from prism-ml, and there's a WebGPU demo in the webml-community space named ternary-bonsai-2-hermes-webgpu. The links were submitted by /u/xenovatech. The source doesn't include further details about model sizes, speed, or what the demo does.
Key Points
- Ternary Bonsai 2 comes in three sizes (2B, 4B, 9B) and is free to download from Hugging Face.
- The smallest model runs in your web browser at up to 500 tokens per second on a MacBook, with no internet needed.
- This means more privacy, no cloud fees, and offline AI—but you'll need a reasonably powerful computer.
Why It Matters
You could soon have fast, private AI on your own device, saving money and keeping your data secure.