Setting up Ollama on Python workbench and UU VRE workspaces

Ollama is a tool that allows large language models to run locally on a workspace. Setting it up in a Python Desktop Workbench or VRE Lab workspace makes it possible to use AI models directly from Python and Jupyter notebooks. You can use Ollama by pulling models to your workspace to automate several tasks, chat with several models (in parallel) without worrying about API keys.

Dependencies

This tutorial assumes you have an installation of uv (Python package manager) on your workspace. Most SURF Research Cloud workspaces now come with uv preinstalled. If (in the rare case) you are using a workspace without any components, then you can install uv yourself with:

curl -LsSf https://astral.sh/uv/install.sh | sh

This tutorial also assumes you have a storage volume attached to your workspace. See the Getting started page for more info about how and why to create a storage volume.

NotePlanning Your Workspace

Before creating your workspace, consider:

  1. Which Ollama model do you want to use?
  2. How much storage do you need? Models can be very large (1GB - 40GB+)
  3. Do you need GPU acceleration? Required for large models (>8B parameters)

Create a virtual environment

Open a terminal

  • In the UU VRE Lab workspace To open a new terminal, click the + button in the file browser and select the terminal in the new Launcher tab
  • Python Workbench, click applications in the topright, and click terminal

Create a project folder

cd data/<the name of your storage volume>
mkdir project
cd project

Install Ollama

curl -fsSL https://ollama.com/install.sh | sh

Create a virtual environment

uv venv --python 3.12
source .venv/bin/activate
uv pip install ipykernel
uv pip install ollama

Pull a model

A few popular models with their resource requirements can be found below:

Model Size Parameters CPU/GPU Best For
qwen2.5:0.5b ~0.4 GB 0.5B CPU Very fast responses, testing, simple coding
llama3.2:1b ~1 GB 1B CPU Multilingual knowledge retrieval, simple tasks
gemma2:2b ~1.6 GB 2B CPU Text generation, natural language processing research
qwen2.5:3b ~2 GB 3B CPU Coding and mathematics, data analysis
llama3.2:3b ~2 GB 3B CPU Summarisation, prompt rewriting, tool use
mistral:7b ~4 GB 7B CPU/GPU Reasoning, text generation, comprehension
llama3.1:8b ~5 GB 8B GPU recommended Coding, complex tasks
qwen2.5:14b ~9 GB 14B GPU recommended Advanced reasoning and coding
gpt-oss:20b ~12 GB 20B GPU (A10) Powerful reasoning, agentic tasks, fine-tunable

Find all available Ollama models here: List of Ollama models

To pull a model, use the following command in the terminal, replacing the model name with the one you want to use.

ollama pull qwen2.5:3b

Create a jupyter kernel

python -m ipykernel install --user --name ollama --display-name "Ollama"

Create a new notebook and use

  • In UU VRE Lab: Click the + button and select the notebook with “Ollama” kernel in the new Launcher tab.

  • In Python Workbench Desktop, run the following command in the terminal to start Jupyter Lab, and create a new notebook with the “Ollama” kernel:

    uv pip install jupyter
jupyter lab

Copy this example code into your notebook, if needed change the model name to the model that you have downloaded and run the cell:

from ollama import chat
from ollama import ChatResponse

response: ChatResponse = chat(model='qwen2.5:3b', messages=[
  {
    'role': 'user',
    'content': 'Why is the sky blue?',
  },
])
print(response['message']['content'])
# or access fields directly from the response object
print(response.message.content)

It may take a few seconds for the first response to come back, but once the model is loaded, you should see a response in your notebook!

LLMs are often used to extract information from documents or images into structured output. As a final exercise we suggest to run the two examples in this blogpost on how to use Ollama for data extraction and image description: Structured Outputs.