Examples¶

Following are examples to train a Deep Learning model on MNIST Data to recognize digits in images using TensorFlow.

Using the Decorator¶

This is the simplest way to use the CodeCarbon tracker with two lines of code. You just need to copy-paste from codecarbon import track_emissions and add the @track_emissions decorator to your training function. The emissions will be tracked automatically and printed at the end of the training.

But you can't get them in your code, see the Context Manager section below for that.

import tensorflow as tf
from codecarbon import track_emissions


@track_emissions(project_name="mnist")
def train_model():
    mnist = tf.keras.datasets.mnist
    (x_train, y_train), (x_test, y_test) = mnist.load_data()
    x_train, x_test = x_train / 255.0, x_test / 255.0
    model = tf.keras.models.Sequential(
        [
            tf.keras.layers.Flatten(input_shape=(28, 28)),
            tf.keras.layers.Dense(128, activation="relu"),
            tf.keras.layers.Dropout(0.2),
            tf.keras.layers.Dense(10),
        ]
    )
    loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)

    model.compile(optimizer="adam", loss=loss_fn, metrics=["accuracy"])

    model.fit(x_train, y_train, epochs=10)

    return model


if __name__ == "__main__":
    model = train_model()

Using the Context Manager¶

We think this is the best way to use CodeCarbon. Still only two lines of code, and you can get the emissions in your code.

import tensorflow as tf

from codecarbon import EmissionsTracker

mnist = tf.keras.datasets.mnist

(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0


model = tf.keras.models.Sequential(
    [
        tf.keras.layers.Flatten(input_shape=(28, 28)),
        tf.keras.layers.Dense(128, activation="relu"),
        tf.keras.layers.Dropout(0.2),
        tf.keras.layers.Dense(10),
    ]
)

loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)

with EmissionsTracker() as tracker:
    model.compile(optimizer="adam", loss=loss_fn, metrics=["accuracy"])
    model.fit(x_train, y_train, epochs=10)

# Display the emissions data
print(f"\nCarbon emissions from computation: {tracker.final_emissions * 1000:.4f} g CO2eq")
print("\nDetailed emissions data:", tracker.final_emissions_data)

Using the Explicit Object¶

This is the recommended way to use the CodeCarbon tracker in a Notebook: you instantiate the tracker and call the start() method at the beginning of the Notebook. You call the stop() method at the end of the Notebook to stop the tracker and get the emissions.

If not in an interactive Notebook, always use a try...finally block to ensure that the tracker is stopped even if an error occurs during training. This is important to ensure the CodeCarbon scheduler is stopped. If you don't use try...finally, the scheduler will continue running in the background after your computation code has crashed, so your program will never finish.

import tensorflow as tf

from codecarbon import EmissionsTracker

mnist = tf.keras.datasets.mnist

(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0


model = tf.keras.models.Sequential(
    [
        tf.keras.layers.Flatten(input_shape=(28, 28)),
        tf.keras.layers.Dense(128, activation="relu"),
        tf.keras.layers.Dropout(0.2),
        tf.keras.layers.Dense(10),
    ]
)

loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)

model.compile(optimizer="adam", loss=loss_fn, metrics=["accuracy"])

tracker = EmissionsTracker()
tracker.start()
try:
    model.fit(x_train, y_train, epochs=10)
except Exception as e:
    print(f"An error occurred: {e}")
finally:
    emissions: float = tracker.stop()
    print(emissions)

Other examples are available in the project GitHub repository.