triton client怎么调用没有config的model

如果您要使用没有配置文件的 Triton Inference Server 模型，则需要使用 Triton Python客户端API来调用模型。您可以使用以下代码片段来加载模型并执行推理：

import numpy as np
import tritonhttpclient
from tritonclientutils import InferenceServerException

# Define server parameters
SERVER_URL = "localhost:8000"
MODEL_NAME = "your_model_name"
MODEL_VERSION = -1 # Use the latest version

# Create a Triton client
triton_client = tritonhttpclient.InferenceServerClient(
    url=SERVER_URL,
    verbose=False,
    concurrency=1
)

# Create a request for the model
request = triton_client.InferRequest()
request.set_model_name(MODEL_NAME)
request.set_model_version(MODEL_VERSION)

# Prepare input data
input_data = np.random.rand(1, 3).astype(np.float32)
request.set_data("input", input_data)

# Send inference request to the server
try:
    response = triton_client.infer(request)
    output_data = response.as_numpy("output")
    print("Model output: ", output_data)
except InferenceServerException as e:
    print("Inference failed: ", e)

请注意，您需要将 SERVER_URL 和 MODEL_NAME 替换为您自己的服务器URL和模型名称。此代码假定您的模型具有名为“input”的输入和名为“output”的输出。