1  from mlflow.deployments import get_deploy_client
 2  
 3  
 4  def main():
 5      client = get_deploy_client("http://localhost:7000")
 6  
 7      print(f"Hugging Face TGI endpoints: {client.list_endpoints()}\n")
 8      print(
 9          f"Hugging Face completions endpoint info: {client.get_endpoint(endpoint='completions')}\n"
10      )
11  
12      # Completions request
13      response_completions = client.predict(
14          endpoint="completions",
15          inputs={
16              "prompt": ("What is Deep Learning?"),
17              "temperature": 0.1,
18          },
19      )
20  
21      print(f"Hugging Face TGI response for completions: {response_completions}")
22  
23  
24  if __name__ == "__main__":
25      main()