Hailo
builds NPU accelerators sold as the AI HAT+ and AI HAT+ 2 for the Raspberry Pi 5. On the software side there are official packages (either from the Raspberry Pi OS apt repository or straight from the vendor), but bringing them up is a sequence of manual steps. With today’s tag of head1328.hailo and head1328.hailo_ollama in version 0.1.0, the two base steps are now available as Ansible roles.
head1328.hailo#
Installs the driver and runtime from the Raspberry Pi OS apt repository. Supported hardware:
| Device | Variant | TOPS |
|---|---|---|
| AI HAT+ (Hailo-8L) | h8l | 13 |
| AI HAT+ (Hailo-8) | h8 | 26 |
| AI HAT+ 2 (Hailo-10H) | h10 | 40 |
Pick the variant via hailo_variant. That’s all 0.1.0 needs - no vendor sources, no source switching. Molecule scenarios are default and simulate.
ansible-galaxy role install head1328.hailohead1328.hailo_ollama#
Installs the hailo-ollama package from the Hailo GenAI Model Zoo and sets up a systemd service that runs Ollama against the Hailo-10H NPU. The Pi then answers on the regular Ollama API - except inference happens on the NPU instead of the CPU.
ansible-galaxy role install head1328.hailo_ollamaIn 0.1.0 this is a lean setup: hailo-ollama as a systemd service, no Open WebUI, no Caddy, no iptables. The “ready to use” variant with a chat frontend and TLS lives in version 0.2.0 , released the day after.
What It’s For#
The fastest bridge between “Pi 5 with AI HAT+ 2 just arrived” and “the Pi serves an Ollama API”. Useful for anyone who wants to bring up the NPU cleanly first, without juggling Open WebUI, Caddy and iptables in the same step.
License#
Both roles: AGPL-3.0-or-later.
Acknowledgments#
- Hailo - NPU hardware and SDK
- Ollama - LLM server
- Ansible Galaxy - role registry
- Molecule - testing framework for Ansible roles
Found this helpful?
Consider supporting via: