A Kubernetes Device Plugin for Hailo AI accelerators, enabling seamless scheduling of AI inference workloads on edge devices like the Raspberry Pi AI HAT+.

The plugin only deploys on nodes with this label, preventing unnecessary pods on nodes without Hailo devices:

kubectl label nodes <node-name> hailo.ai/device=present

kubectl apply -f https://raw.githubusercontent.com/gllm-dev/hailo-device-plugin/main/deploy/daemonset.yaml

# Check pod is running
kubectl -n kube-system get pods -l app.kubernetes.io/name=hailo-device-plugin

# Check device is registered
kubectl get nodes -o custom-columns=NAME:.metadata.name,HAILO:.status.allocatable.hailo\\.ai/h10

Should show:

NAME          HAILO
<node-name>   1

cat << 'EOF' | kubectl apply -f -
apiVersion: v1
kind: Pod
metadata:
  name: hailo-test
spec:
  containers:
  - name: test
    image: debian:bookworm-slim
    command: ["sh", "-c", "ls -la /dev/hailo* && sleep 3600"]
    resources:
      limits:
        hailo.ai/h10: 1
  restartPolicy: Never
EOF

# Check logs
kubectl logs hailo-test

Should show /dev/hailo0

apiVersion: v1
kind: Pod
metadata:
  name: inference
spec:
  containers:
    - name: model
      image: your-inference-image
      resources:
        limits:
          hailo.ai/h10: 1

• Kubernetes v1.26+
• Hailo driver installed on worker nodes (installation guide)
• Hailo device accessible at /dev/hailo*

The DaemonSet uses a node selector (hailo.ai/device=present) to deploy only on labeled nodes.

1. Label nodes with Hailo devices:

kubectl label nodes <node-name> hailo.ai/device=present

2. Deploy the plugin:

kubectl apply -f https://raw.githubusercontent.com/gllm-dev/hailo-device-plugin/main/deploy/daemonset.yaml

Configure the plugin using environment variables in the DaemonSet:

Edit the DaemonSet to add environment variables:

kubectl edit daemonset hailo-device-plugin -n kube-system

spec:
  template:
    spec:
      containers:
        - name: hailo-device-plugin
          env:
            - name: HAILO_RESOURCE_NAME
              value: "hailo.ai/h8l"
            - name: HAILO_ARCHITECTURE
              value: "HAILO8L"

This plugin handles device scheduling and allocation. For inference workloads, you'll need to build containers with HailoRT on your Hailo nodes (the SDK is not publicly redistributable).

Resources:

• HailoRT SDK - Runtime library for Hailo devices
• Hailo Model Zoo - Pre-trained models and examples
• hailo_model_zoo_genai - LLM support (Ollama-compatible API)

git clone https://github.com/gllm-dev/hailo-device-plugin.git
cd hailo-device-plugin

go build -o hailo-device-plugin ./cmd/plugin

docker build -t hailo-device-plugin:local .

Contributions are welcome! Please read our Contributing Guidelines before submitting a Pull Request.

This project follows Conventional Commits and Keep a Changelog.

This project is licensed under the MIT License - see the LICENSE file for details.

• Hailo for their AI accelerator technology
• Kubernetes Device Plugin framework