Deepseek Pi

Using Ansible and Docker Compose, run Ollama with Open WebUI to host and interact with models such as Deepseek R1. This is just a bit of fun, and isn't really usable in real-world scenarios. For comparison, the Deepseek R1 that at the time of writing is supposedly beating ChatGPT runs with 671 billion parameters, whereas the the only version that works on a standard Pi uses 1.5 billion parameters.

Setup

Inventory File

Configure inventory.yaml with the details of the target machine.

SSH

Run the following, following the onscreen instructions where necerssary:

ssh-keygen
ssh-copy-id <remote-user>@<remote-host>
ansible -i inventory.yaml all -m ping

This should return a pong. Now run:

ansible-playbook -i inventory.yaml playbook.yaml

Usage

Navigate to http://<remote-host>:3000 and create an admin login. Now navigate to http://<remote-host>:3000/admin/settings/ and click on models and manage models (looks like a small download link). In the Pull a model from Ollama.com box enter deepseek-r1:1.5b or whatever model you want to run.