Inference with local models

Learn how to run inference using locally stored models on a DeGirum AI Server, whether the server runs on the same machine as the client or remotely over the network.

Estimated read time: 3 minutes

This setup runs inference on local hardware using DeGirum AI Server. The model is stored in a local folder on the machine where the server runs.

You can set the server address via inference_host_address="localhost" or inference_host_address="<host_ip>:<port>".

Client and AI Server on the same host

In this case, both the AI Server and your client application run on the same machine.

First, download the model to a local folder using the degirum download-zoo command:

ZOO="$HOME/degirum_model_zoo"
mkdir -p "$ZOO"

degirum download-zoo \
  --path "$ZOO" \
  --url https://hub.degirum.com/degirum/hailo \
  --model_family yolov8n_coco--640x640_quant_hailort_multidevice_1

Then launch the AI Server with the following command:

degirum server --zoo "$ZOO"

You should see output like:

DeGirum asio server is started at TCP port 8778
Local model zoo is served from '/home/degirum/degirum_model_zoo' directory.
Press Enter to stop the server

Zoo path may differ.

The server runs until you press ENTER. By default, it listens on TCP port 8778. To change the port, use the --port argument:

degirum server --port <your_port> --zoo "$ZOO"

Example ModelSpec

This example configures ModelSpec to use the AI Server and load the model from the local zoo:

# Example ModelSpec
model_spec = ModelSpec(
    model_name="yolov8n_coco--640x640_quant_hailort_multidevice_1",
    zoo_url="aiserver://",
    inference_host_address="localhost",
    model_properties={"device_type": ["HAILORT/HAILO8L", "HAILORT/HAILO8"]}
)

• zoo_url="aiserver://": load the model from the zoo path given when launching the AI Server.

• inference_host_address="localhost": run inference using the Hailo device managed by the local server.

Client and AI Server on different hosts

To run inference from a separate client machine, first use degirum download-zoo to download the model to the host that will run the AI Server:

ZOO="$HOME/degirum_model_zoo"
mkdir -p "$ZOO"

degirum download-zoo \
  --path "$ZOO" \
  --url https://hub.degirum.com/degirum/hailo \
  --model_family yolov8n_coco--640x640_quant_hailort_multidevice_1

Star the AI Server on the remote host:

degirum server --zoo "$ZOO"

You should see output like:

DeGirum asio server is started at TCP port 8778
Local model zoo is served from '/home/degirum/degirum_model_zoo' directory.
Press Enter to stop the server

As before, the server runs until you press ENTER. By default, it listens on TCP port 8778. To change the port, use the --port argument:

degirum server --port <your_port> --zoo "$ZOO"

Example ModelSpec

This time, the client points to a remote host:

# Example ModelSpec
model_spec = ModelSpec(
    model_name="yolov8n_coco--640x640_quant_hailort_multidevice_1",
    zoo_url="aiserver://",
    inference_host_address="<host_ip>:<port>",
    model_properties={"device_type": ["HAILORT/HAILO8L", "HAILORT/HAILO8"]}
)

• zoo_url="aiserver://": still points to the AI Server zoo.

• inference_host_address="<host_ip>:<port>": runs inference using the Hailo device on the remote AI Server.