我想使用tensorrt8来推理onnx模型，请给出代码

好的，以下是使用TensorRT 8对ONNX模型进行推理的Python代码示例：

import tensorrt as trt
import numpy as np

# Load ONNX model
onnx_model_path = "/path/to/onnx/model.onnx"
onnx_encoder = trt.OnnxParser(trt.Logger(trt.Logger.WARNING))
onnx_encoder.parse(open(onnx_model_path, 'rb').read())

# Set TensorRT builder parameters
builder = trt.Builder(trt.Logger(trt.Logger.WARNING))
builder.max_workspace_size = 1 << 20
builder.max_batch_size = 1

# Set input and output shapes
input_shape = (1, 3, 224, 224)
output_shape = (1, 1000)
builder.fp16_mode = True
builder.strict_type_constraints = True

network = builder.create_network()
input_tensor = network.add_input(name="input", dtype=trt.float32, shape=input_shape)
output_tensor = network.add_output(name="output", dtype=trt.float32, shape=output_shape)

# Add ONNX model to TensorRT network
onnx_tensor = {}
for i in range(onnx_encoder.num_outputs):
    onnx_tensor[onnx_encoder.get_output_name(i)] = network.add_output(
        name=onnx_encoder.get_output_name(i),
        dtype=trt.float32,
        shape=onnx_encoder.get_output_shape(i))

onnx_tensor.update(
    {onnx_encoder.get_input_name(i): network.add_input(
        name=onnx_encoder.get_input_name(i),
        dtype=trt.float32,
        shape=onnx_encoder.get_input_shape(i))
        for i in range(onnx_encoder.num_inputs)})

# Build TensorRT engine
engine = builder.build_cuda_engine(network)
context = engine.create_execution_context()

# Generate random input data
input_data = np.random.rand(*input_shape).astype(np.float32)

# Allocate device memory for input and output data
d_input = cuda.mem_alloc(input_data.nbytes)
d_output = cuda.mem_alloc(output_shape[0] * np.dtype(np.float32).itemsize)

# Copy input data to device memory
cuda.memcpy_htod(d_input, input_data)

# Do inference
context.execute(1, [int(d_input), int(d_output)])

# Copy output data from device memory
output_data = np.empty(output_shape, dtype=np.float32)
cuda.memcpy_dtoh(output_data, d_output)

# Print output data
print(output_data)

请注意，此示例假定您已经正确安装了TensorRT 8并且已经添加了必要的路径。如果您需要更多帮助，请参阅TensorRT文档和示例。