Skip to main content
NVIDIA
NVIDIA

cuopt

Downloadable

World-record accuracy and performance for complex route optimization.

Get API Key
API Reference
Accelerated by DGX Cloud
Deploying your application in production? Get started with a 90-day evaluation of NVIDIA AI Enterprise

Follow the steps below to download and run the NVIDIA NIM inference microservice for this model on your infrastructure of choice.

Generate API Key

Pull and Run the NIM

  1. Export NGC_API_KEY variable.
export NGC_API_KEY=<PASTE_API_KEY_HERE>
  1. Run the NIM container with the following commands.
$ docker login nvcr.io
Username: $oauthtoken
Password: <PASTE_API_KEY_HERE>

For CUDA 12.9 and Python 3.13:

docker run -it \
    --gpus='"device=0"' \
    -p 5000:5000 \
    -e NGC_API_KEY \
    nvcr.io/nvidia/cuopt/cuopt:latest-cuda12.9-py3.13

For CUDA 13.0 and Python 3.13:

docker run -it \
    --gpus='"device=0"' \
    -p 5000:5000 \
    -e NGC_API_KEY \
    nvcr.io/nvidia/cuopt/cuopt:latest-cuda13.0-py3.13

This command will start the NIM container and expose port 5000 for the user to interact with the NIM.

  1. Open a new terminal, leaving the terminal open with the just launched service. In the new terminal, wait until the health check end point returns {"status":"RUNNING","version":"<VERSION>"} before proceeding. This may take a couple of minutes. You can use the following command to query the health check.
curl http://localhost:5000/v2/health/ready

Test the NIM

Python client example

  1. Save following Python example to a file named nim_client.py.
#!/usr/bin/env python3
import requests
import time

data = {"cost_matrix_data": {"data": {"0": [[0,1],[1,0]]}},
           "task_data": {"task_locations": [0,1]},
           "fleet_data": {"vehicle_locations": [[0,0],[0,0]]}}

response = requests.post(
    url="http://localhost:5000/cuopt/request",
    json=data
)
response_body = response.json()

poll_interval = 2  # Time in seconds between polls

solution_url = "http://localhost:5000/cuopt/solution/" + response_body["reqId"]

while True:
    response = requests.get(solution_url)
    if response.status_code == 200 and "response" in response.json().keys():
        print(response.json())
        break
    elif response.status_code == 200:
        print(f"Polling for response")
    else:
        response.raise_for_status()
    time.sleep(poll_interval)
  1. Execute the example.
chmod +x nim_client.py

./nim_client.py

Shell client example

  1. Save the following Shell example to a file named nim_client.sh.
#!/usr/bin/env bash
set -e

pip install jq

REQUEST_URL=http://localhost:5000/cuopt/request
SOLUTION_URL=http://localhost:5000/cuopt/solution

data='{"cost_matrix_data": {"data": {"0": [[0,1],[1,0]]}},
           "task_data": {"task_locations": [0,1]},
           "fleet_data": {"vehicle_locations": [[0,0],[0,0]]}}'
reqId=$(curl -H 'Content-Type: application/json' \
        -d "$data" "$REQUEST_URL" | jq -r '.reqId'
       )

echo $reqId

while true; do
  response=$(curl -s "$SOLUTION_URL/$reqId")

  if echo "$response" | jq -e '.response.solver_response.status == 0' > /dev/null; then
    echo "$response"
    break
  fi

  echo "Polling for response"

  sleep 2

  done
  1. Execute the example.
chmod +x nim_client.sh

./nim_client.sh
  1. The NIM displays the results to the terminal in JSON format.

For more details on getting started with this NIM, visit the NVIDIA CUOPT Docs