Environment

This page takes about 10 minutes. You install uv, a tool that sets up Python and the packages for you. Then you get the notebooks, start JupyterLab, run one notebook that checks everything, and download two index files for notebook 03. The page assumes that you have already installed Ollama and downloaded the models (see Install Ollama and Download the models).

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

git clone https://github.com/nishad/llm-workshop-notebooks.git
cd llm-workshop-notebooks

uv sync

uv run jupyter lab

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

git clone https://github.com/nishad/llm-workshop-notebooks.git
cd llm-workshop-notebooks

uv sync

uv run jupyter lab

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
jupyter lab

py -3.12 -m venv .venv
.venv\Scripts\Activate.ps1
pip install -r requirements.txt
jupyter lab

Download the prebuilt index files

Notebook 03 builds search indexes for two vocabularies, STW and UNESCO. On an Apple Silicon Mac with 64 GB, building them takes about seven minutes: five for STW and two for UNESCO. On a CPU-only laptop it takes far longer, so download the prebuilt files instead. On an 8 GB machine this download is the only way to follow notebook 03 live. The download link is in step 4 of Before you start. Put both files in the runs/ folder of the repository (create the folder if it does not exist). The notebook then loads the files instead of building them.

To check that a file downloaded correctly, run shasum -a 256 runs/stw-de-labels.sqlite on macOS and Linux, or Get-FileHash runs\stw-de-labels.sqlite in PowerShell. The result should match the sha256 column in the table.

VS Code instead of JupyterLab

Open the cloned folder in VS Code and select the .venv interpreter as the kernel. You can then run the notebooks there and skip uv run jupyter lab.

Windows notes

• Use PowerShell, not Command Prompt. The commands above are written for it.
• The py launcher picks a Python version: py -3.12 runs Python 3.12. It is useful when several versions are installed.
• Clone into a short path such as C:\src to avoid long path names.
• If PowerShell refuses to run the activation script, use the uv steps instead. They do not need that script.

What a ready machine shows

The readiness report is cell 00.02 of notebook 00. This is how it looks on a current install, with every line green.

✅ Python 3.12 or newer: found 3.12.12
✅ Memory: 64 GB, use nishad/apertus:8b (16 GB tier)
✅ Free disk 20 GB: 635 GB free
✅ Ollama 0.40 or newer: version 0.40.4 at http://localhost:11434
✅ Model nishad/apertus:8b: present
✅ Chat template (Unsloth fixes): present
✅ Embedding model qwen3-embedding:0.6b: 1024 dimensions
✅ Smoke chat: model said: 'OK'
✅ Smoke tool call: proposed lookup_authority({'name': 'Hannah Arendt'})

All good.

This output comes from notebook 00, run on Ollama 0.40.4 on 15 October 2026.

Your memory and disk numbers will differ. The Memory line tells you which tier you are on, and so which model to use: 16 GB or more uses nishad/apertus:8b, 8 GB uses nishad/apertus:4b. On an 8 GB laptop, set the environment variable WORKSHOP_TIER=8gb before you start JupyterLab, or change "16gb" to "8gb" in the first cell of each notebook. The tier-8gb/ folder of the repository also holds notebooks 04 and 05 with the outputs of a run on the 8 GB tier, so you can see what the 4B model does before you run it.

If a line is red, follow the hint in it and run the cell again. Paste the output into the shared document when you ask for help. More fixes are on Troubleshooting.