Pull and run your first model

Start with a small model so the first experiment works on modest hardware.

This course uses Gemma 3 1B:

ollama run gemma3:1b

The first run downloads an approximately 815 MB Q4_K_M artifact. Later runs use the local copy. Watch the progress bar once so you know what “first pull” feels like on your connection.

At the prompt, enter:

Explain what a model weight is in two sentences.

You should get a short plain-language answer about learned numbers that drive predictions. It will not match this lesson word for word. That is fine. We are checking the pipeline, not grading prose.

Leave the chat with:

/bye

List stored models:

ollama list

You should see gemma3:1b with a size near 815 MB and a recent modified time.

The 1B model is useful for checking the complete pipeline. Do not use its successful installation as evidence that it can handle every application task. We will evaluate quality separately with real inputs.

If your machine has more memory, you can later compare a larger model using the same evaluation examples. Keep the tag in your notes: gemma3:1b is a specific quantized artifact, not a vague “Gemma” label.

When ollama run fails with a pull error, check disk space and network first. When it fails after pull, check memory and quit other heavy apps. The error message usually points at one of those two buckets.

If you prefer a non-interactive smoke test, use the API instead:

curl -s http://localhost:11434/api/chat -d '{
  "model": "gemma3:1b",
  "messages": [{"role": "user", "content": "Reply with the word ready."}],
  "stream": false
}'

You should see "ready" or a similar short token inside message.content.

Try this before you continue: run the prompt above and confirm ollama list shows the model. If that works, your runtime and weights path is alive.