triton_http_client_basic_model_inference_quickstart.py

1import numpy as np
2import tritonclient.http as httpclient
3
4# Initialize the client
5# Assumes Triton is running locally on default port 8000
6try:
7    triton_client = httpclient.InferenceServerClient(url="localhost:8000")
8except Exception as e:
9    print("context creation failed: " + str(e))
10    exit(1)
11
12# Define the model name and input data
13model_name = "simple"
14input_data = np.arange(16, dtype=np.int32).reshape(1, 16)
15
16# Setup inputs and outputs
17inputs = []
18outputs = []
19inputs.append(httpclient.InferInput('INPUT0', [1, 16], "INT32"))
20
21# Initialize the data
22inputs[0].set_data_from_numpy(input_data)
23
24outputs.append(httpclient.InferRequestedOutput('OUTPUT0'))
25
26# Test with HTTP
27results = triton_client.infer(model_name, inputs, outputs=outputs)
28
29# Get the output as a numpy array
30output0_data = results.as_numpy('OUTPUT0')
31
32print(output0_data)