Preview · Open model
Kisoku, a language model we trained from scratch.
1.6 billion parameters, free and open. It runs offline on your own laptop, so nothing you type leaves your computer.
This is a preview. The full release lands in October.
- Size
- 1.6 billion parameters
- Download
- 1.0 to 3.2 GB
- Speed
- About 150 words a second on a MacBook Pro with an M5 Max chip
- Trained on
- About 417 billion tokens of text (a token is roughly three quarters of a word)
- Language
- English
- License
- MIT for the weights (details on the model page)
Run it in three steps
Kisoku runs in Ollama, a free app that runs AI models on your own computer. The commands below are for the Terminal app on a Mac, or a Linux terminal.
Download Kisoku
This grabs the recommended 1.7 GB version and its setup file into a new folder.
$ mkdir kisoku && cd kisoku$ curl -LO https://huggingface.co/0arch-io/kisoku-1.6b-preview-GGUF/resolve/main/kisoku-1.6b-preview-Q8_0.gguf$ curl -LO https://huggingface.co/0arch-io/kisoku-1.6b-preview-GGUF/resolve/main/Modelfile
Start chatting
The first command sets it up, the second opens a chat. Type /bye to leave.
$ ollama create kisoku -f Modelfile$ ollama run kisoku
What it does well
- Everyday questions and conversation.
- Short writing: emails, summaries, rewrites, poems.
- Simple code, like a small Python function with an example.
- Answers directly instead of lecturing or refusing.
Where the preview is rough
- Math and multi-step reasoning. It can get simple word problems wrong.
- Obscure facts. It can sound confident when it is wrong, so check anything that matters.
- It has no internet access and knows nothing about current events.
- It remembers about 3,000 words of a conversation at a time.
Coming in October
The preview is the first of three training stages. The full release adds:
- 01
More math and code training, aimed squarely at the reasoning gaps above.
- 02
Saying “I don’t know” instead of making something up, and owning a mistake when you correct it.
- 03
Tool use, so it can call things like web search instead of guessing.