Unit 0. Welcome to the course

A local model, before you need one

Optional, free, and worth doing before the session that asks for it rather than during it. Nothing scored in this course needs a model at all — every check runs offline on fake. This page is for the sessions that ask you to compare one model to another.

Why bother

• No key, no card, no bill. The model runs on your machine and nothing leaves it
• It works offline. On a plane, on hotel wifi, with the network unplugged
• It is the same code path. One line of .env changes the lane; your notebook does not change
• The alternative costs money. A key made in a provider console is billed per token — Claude Pro does not include API credits

What it costs you

• Money — nothing, ever
• Disk — 4.7 GB for the model this course names
• RAM — about 8 GB while it is answering
• Network — the download, once, and then none at all
• Patience — the first answer after a restart is slow, because the model is being read into memory. After that it is quick

Install it

• Download from https://ollama.com/download — a .dmg on macOS, an .exe on Windows, one line on Linux
• Check it arrived — ollama --version
• Linux without root? The manual install works: unpack the release tarball into ~/.local and run ollama serve yourself

Pull the model

• ollama pull qwen2.5:7b-instruct
• 4.7 GB the first time. Start it before dinner, not before class
• Confirm it landed — ollama list
• Tight on RAM? ollama pull qwen2.5:3b-instruct is about half the size. Set it as BOOTCAMP_MODEL and everything else is the same

Point the course at it

• Open .env in the course folder — bootcamp start already wrote it
• Change one line — BOOTCAMP_PROVIDER=ollama
• If you chose a different model, set BOOTCAMP_MODEL= to its name
• Check it — uv run bootcamp doctor

You want these three:

✅ Ollama server reachable at http://localhost:11434/v1
✅ Ollama model qwen2.5:7b-instruct pulled
✅ Ollama round-trip (one short completion)

The third one is the one that matters. The doctor asked your model for a completion and got one back.

When it does not work

• no server at http://localhost:11434/v1 — Ollama is installed but not running. ollama serve in another terminal, or start the desktop app
• model ... not pulled — the pull did not finish. Run it again; it resumes
• It answers, very slowly — that is a 7B model on a CPU. Try the 3B, or go back to fake
• Anything else — uv run bootcamp doctor names the fix. Post its output in the group, never your .env

Going back

• BOOTCAMP_PROVIDER=fake in .env, and you are on the offline lane again
• Every scored exercise passes there. Switching back costs you no marks and never has