← Course overview

Lesson 3 of 8 · 25 minutes

In this lesson

Open Weight Models · Lesson 3

Choose and run your first model

Download a small model, start a chat, and inspect the model running on your computer.

What you will learn

  • Inspect a model's identity, licence, and input capabilities before using it.
  • Pull gemma3:1b, start a conversation, and send a simple one-shot request.
  • Distinguish downloaded models from models currently loaded in memory.
  • Exit a chat and unload a model without confusing either action with deletion.

Choose a specific model, not just a family name

Open the gemma3:1b library entry and Google's Gemma 3 model card before downloading. The text before the colon identifies the model family; the tag chooses a particular variant. Use the exact course tag throughout the first exercises so that comparisons have a consistent starting point.

The 1B variant is text-only, with a published 32K context capability. The library currently lists this download at about 815 MB; recheck the page because packages can change. Do not apply the larger Gemma 3 models' image capability to this variant, or interpret its download size as the amount of RAM it will use.

A small model is useful for learning the workflow but may miss instructions, invent details, or struggle with multi-step tasks. Your first objective is a working, inspectable local run, followed by an honest quality check.

Download once, then inspect the local package

Run the commands below in a terminal after completing the setup checks. Wait for the pull to finish before inspecting the model. It downloads files; it does not train the model. The information and licence views help connect the local package to the model documentation you reviewed.

  • Record the model name and ID from your local inventory.
  • Record the displayed architecture, parameter size, and quantisation where available.
  • Check the licence against the current publisher terms; retain the model card link with your notes.

Download and inspect the course model · text

ollama pull gemma3:1b
ollama list
ollama show gemma3:1b
ollama show gemma3:1b --license

Run your first local chat

Start the model with the terminal command below. At the chat prompt, type: Explain the difference between RAM and disk storage in two short sentences for a beginner. Wait for the answer and check both its content and its requested length.

A response demonstrates that inference completed, not that the explanation is correct. Compare its claims with the distinction from Lesson 1. Try one follow-up asking for a simple example, then type /bye at the model's chat prompt to return to your shell.

Start the interactive chat from your terminal · text

ollama run gemma3:1b

Leave the chat: type this inside Ollama · text

/bye

Send a separate one-shot request from your terminal · text

ollama run gemma3:1b "Define inference in one sentence for a beginner."

Inspect what is loaded, then unload it

While the model is active, open another terminal and inspect ollama ps. Its processor information describes whether the run uses CPU, GPU, or a split. That is an observation about this run, not a performance promise for every model or task.

Leaving a chat and unloading its model are separate actions. The runner can retain a model in memory for later requests. After you finish the chat, use stop and check again. The model should disappear from the running list while remaining in the downloaded inventory.

Inspect what is loaded, then unload it
CommandQuestion it answers
ollama listWhich model packages are stored locally?
ollama show gemma3:1bWhat information describes this package?
ollama psWhich models are loaded now?
ollama stop gemma3:1bCan I unload this model while retaining its files?

Inspect and release the loaded model · text

ollama ps
ollama stop gemma3:1b
ollama ps
ollama list

Keep a useful first-run record

Create a small experiment log containing the date, Ollama version, model tag and ID, prompt, complete answer, and one specific observation. A screenshot alone can hide the input that produced an answer. Text records make later comparisons easier.

If the run fails, separate download, service, and resource problems. Check the exact tag for a missing-model error, revisit Lesson 2 for a service error, and inspect the relevant hardware guidance for a memory failure. Do not interpret a long wait as proof of either successful inference or a broken installation.

Expected variation

The sample prompts have no guaranteed exact wording. Judge whether the explanation is accurate and follows its constraints; save mistakes as evidence for the next lesson.

Put it into practice

Complete a model lifecycle

  1. Review the model entry and terms, then pull gemma3:1b. Inspect it and record its local name and ID.

  2. Start an interactive chat. Ask the RAM-versus-disk prompt and save its answer with your accuracy and length checks.

  3. While the chat is active, run ollama ps in a second terminal and record the processor information you actually observe.

  4. Leave the chat with /bye, then unload gemma3:1b with ollama stop.

  5. Compare ollama ps and ollama list after stopping it. Save a one-sentence explanation of why their inventories differ.

You have completed this task when…

  • gemma3:1b appears in the downloaded inventory, and you have captured a completed response.
  • Your log identifies at least one checked claim and whether the answer met the requested two-sentence format.
  • After stopping, the model is no longer loaded but its package is still listed locally.

Official documentation

Use these references for platform requirements, current options, and further detail.

Check your understanding

Choose an answer for each question, then check your reasoning.

1. After ollama stop gemma3:1b, the model remains in ollama list. Why?
2. A larger model in the Gemma 3 family accepts images. What should you assume about gemma3:1b?

Answer each question to continue.