Open Weight Models · linux · Keep nearby

Ollama quick reference · linux

Use these commands from a Linux client shell. Keep the same server mode throughout the course. Commands marked systemd require an installed ollama.service on a machine running systemd; ollama serve is the alternative foreground workflow, not an extra server to start alongside it.

Everyday commands

Command	Use it to…
`cd "$HOME/CyberCorps/open-weight-models"`	Return to the folder containing your Python files, Modelfile, and evidence.
`python3 --version`	Check the Python 3 interpreter used by the API and capstone scripts.
`ollama --version`	Inspect the installed Ollama client version.
`ollama list`	Check the server connection and its downloaded model inventory.
`ollama pull gemma3:1b`	Download the local course model.
`ollama run gemma3:1b`	Open a local model chat; enter /bye inside the chat to return to the shell.
`ollama show gemma3:1b`	Inspect the selected model's details.
`ollama ps`	Inspect loaded models and their observed processor allocation.
`ollama stop gemma3:1b`	Unload the base model while keeping its downloaded files.
`ollama create cybercorps-study -f Modelfile`	Create or update the named assistant from the file in your current project folder.
`ollama show --modelfile cybercorps-study`	Inspect the configuration stored under the capstone's model name.
`python3 local_chat.py`	Run the saved lesson 5 client from the project folder.
`python3 study_assistant.py`	Run the five capstone cases using cybercorps-study.
`free -h`	Inspect system memory where procps/procps-ng free is installed; this does not report all GPU memory.
`systemctl status ollama --no-pager`	Systemd only: inspect the installed Ollama service before deciding whether to start it.
`sudo systemctl start ollama`	Systemd only: start the installed service if it is inactive.
`sudo journalctl -u ollama -b -n 100 --no-pager`	Systemd only: read recent Ollama logs from the current boot.
`sudo systemctl edit ollama`	Systemd only: edit the service override; lesson 7 shows the local-only Environment setting and restart steps.
`OLLAMA_NO_CLOUD=1 ollama serve`	Manual foreground alternative: start only after stopping your own prior foreground server and confirming no competing service; verify its startup output.

Run one command at a time. Model removal changes your local installation; unloading a model only releases its memory. See Lesson 7 for troubleshooting and privacy settings.

Make a local API request

Start Ollama and download gemma3:1b first. These examples request one complete JSON response. The answer is in message.content. Use the example for your shell.

Linux shell: send a local chat request · bash

curl --fail --show-error --max-time 120 \
  http://localhost:11434/api/chat \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "gemma3:1b",
    "stream": false,
    "messages": [
      {"role": "user", "content": "Note: the workshop starts at 09:30. Using only this note, what time does it start?"}
    ]
  }'

The local endpoint needs no API key. Keep it on your own machine; it is not an authenticated public service. See Lesson 5 for a Python client and conversation history.

Key terms

Weights

The learned numerical parameters used by a model. Their availability does not by itself describe the permissions in its licence.

Inference

Using an existing model to generate a result from an input.

Token

A unit of text processed by the model. A word may occupy more than one token.

Context

The working input available for a generation, including instructions, supplied material, and conversation history.

Quantisation

Representing model values with fewer bits to reduce storage and memory needs, with possible quality trade-offs.

Model tag

A model identifier such as gemma3:1b. Record the model ID as well when you need to track a particular downloaded version.

Modelfile

An Ollama configuration that selects a base model and supplies settings or instructions.

Evaluation

Checking outputs against defined tasks and expected behaviour, while recording errors as well as successes.