Skip to content

Ansible Role: bsmeding.ollama

The following documentation is taken from the repository README.

test status

This role deploys Ollama on a Linux server using the official installation script. Ollama runs local LLMs (Llama, Mistral, etc.) for inference.

Requirements

  • Linux (Debian, Ubuntu, RHEL, Fedora, Arch)
  • systemd (for service management)
  • curl, zstd (installed by the role when ollama__install_prerequisites is true)

Role Variables

Available variables are listed below, along with default values (see defaults/main.yml):

Variable Default Description
ollama__skip_setup false When true, setup is skipped.
ollama__version '' Pin Ollama version (e.g., 0.5.7). Empty = latest.
ollama__install_prerequisites true Install curl and zstd before running the install script.
ollama__state started Service state: started, stopped, or restarted.
ollama__enabled true Enable the ollama service at boot.
ollama__host 127.0.0.1 Bind address (use 0.0.0.0 for remote access).
ollama__port 11434 API listening port.
ollama__models [] List of models to pull (e.g., ['llama3.2', 'mistral', 'codellama']).

Example Playbook

- hosts: ollama_servers
  roles:
    - role: bsmeding.ollama
      vars:
        ollama__version: "0.5.7"  # optional: pin version
        ollama__enabled: true
        ollama__port: 11435  # optional: custom port
        ollama__host: 0.0.0.0  # optional: bind to all interfaces for remote access
        ollama__models:
          - llama3.2
          - mistral
          - codellama

For Open WebUI (web interface), use the separate bsmeding.openwebui role.

GPU Support

The official install script automatically detects and configures:

  • NVIDIA GPU: Installs CUDA drivers when detected.
  • AMD GPU: Downloads and installs ROCm-enabled Ollama build.
  • CPU-only: Runs in CPU mode when no GPU is detected.

Post-Install

After deployment, the Ollama API is available at http://<host>:<port> (default: http://127.0.0.1:11434). Use ollama__host and ollama__port to customize. Models listed in ollama__models are pulled automatically; you can also pull additional models manually:

ollama pull llama3.2

License

MIT

Author Information

This role was created by Bart Smeding.