precey/chatnano-models

ChatNano models

These are the three models the ChatNano desktop app runs on your laptop's CPU. The app downloads them itself, so there's nothing here you need to fetch by hand.

File	Model	Converted from	Size
nano-tokyo.nano	Nano Tokyo, 2B	Ternary-Bonsai-2-2B	429 MB
nano-cairo.nano	Nano Cairo, 4B	Ternary-Bonsai-2-4B	938 MB
nano-oslo.nano	Nano Oslo, 9B	Ternary-Bonsai-2-9B	1.99 GB

They're PrismML's Ternary Bonsai 2 models, Qwen3.5 trained so its weights are -1, 0 or +1. We converted them to our own .nano format without changing a single weight. The conversion only repacks them so our engine reads them fast: five ternary digits to a byte, and one f16 scale for every 128 weights.

Nano Tokyo and Nano Oslo were called Nano Nairobi and Nano Lagos when we converted them, and those old names are still written inside the files. The app goes by the file name.

The vision encoders aren't here. ChatNano gets them from PrismML's own repos, unchanged.

License

Apache 2.0, like the models they come from. Created using Bonsai by Prism ML.

PrismML's notice, as published with each model (the 2B and 9B name Qwen3.5-2B and Qwen3.5-9B instead):

This software is copyright 2026-present Prism ML, Inc. It is available under the Apache 2.0 license. If you publicly deploy or redistribute this software, we would appreciate attribution such as: "Created using Bonsai by Prism ML."

This software is built from Qwen3.5-4B, Copyright 2026 Alibaba Cloud, which is available under the Apache 2.0 License: https://huggingface.co/Qwen/Qwen3.5-4B/blob/main/LICENSE