---
license: apache-2.0
base_model:
  - Qwen/Qwen3.5-0.8B
  - Qwen/Qwen3.5-2B
  - Qwen/Qwen3.5-4B
pipeline_tag: text-generation
tags:
  - webgpu
  - webassembly
  - browser
---

# tiny-ai's files

What the [tiny-ai chat](https://huggingface.co/spaces/LilVague/tiny-ai) downloads and runs in your browser: Qwen3.5 at
4 bits in the chat's own format (`models/`), the packages it pulls in by subject (`packs/`), the router that picks
them (`router/`) and their scores (`packages.json`). They are files for that page, not for other libraries.

The models are Qwen's Qwen3.5 (Apache 2.0), changed by us: rounded to 4 bits, their vocabulary cut to the tokens
English and code use, and some of their weights edited. Also at [tiny-ai.duckii.dev](https://tiny-ai.duckii.dev).