Executor¶

Executor realizes a simple executor in which all operators will be executed in order. You can run Executor in a Python script.

The logic of Executor is very simple. It is suggested to thoroughly run the model with Executor in debugging phase on one computer and then switch to mode of multiple devices or multiple computers to compute.

Executor accepts either a Place including CPUPlace and CUDAPlace, or a device descriptor str such as "cpu" and "gpu:0".

         import paddle
import numpy as np

# Enable static graph mode and configure compute device.
paddle.enable_static()
place = (
    paddle.CUDAPlace(0)
    if "gpu:0" in paddle.device.get_available_device()
    else paddle.CPUPlace()
)

# First create the Executor to run computation graph.
exe = paddle.static.Executor(place)

# define input data and labels
# shape '-1' indicates dynamic batch size and '784' is the feature dimension
x = paddle.static.data(name="x", shape=[-1, 784], dtype="float32")
label = paddle.static.data(name="label", shape=[-1, 1], dtype="int64")

# define a fully connected layer with an output dimension of 10, and cross entropy loss function
out = paddle.static.nn.fc(name="fc", x=x, size=10, activation="relu")
loss = paddle.nn.functional.cross_entropy(out, label)

# Run the startup program once to initialize parameters.
exe.run(paddle.static.default_startup_program())

# create a 1x784 matrix filled with 1 as input data, and a label of 0 (1x1 matrix)
feed_dict = {
    "x": np.ones((1, 784), dtype="float32"),
    "label": np.array([[0]], dtype="int64"),
}

# Run the main program using feed_dict as input and get the output of loss and out.
(loss_value, out_value) = exe.run(
    paddle.static.default_main_program(), feed=feed_dict, fetch_list=[loss, out]
)
print(f"loss: {loss_value}\nmodel output: {out_value}")

        

For the parameter introduction about Executor please refer to Executor