MLXChallenge

In MLX, we are data lovers and algorithm trainers, but we do not have any tool to be aware of a training energy consumption. Code Carbon brings a new method to evaluate your consumption per training, and save it across time. This small exercise will help you to use this simple tool.

A second part is dedicated to front lovers (#AUMA), to evaluate website consumption thanks to GreenIT tool.

1 - Which consommation for a training ?

The aim of this part, is to measure environment impact during algorithm training

• First you'll need to install necessary depedencies pip install -r requirements.txt

• We will work on mnist dataset that you can load with

mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()

• Choose an algithm to predict correctly digits

• Now estimate the emission of your training (add the fit between tracker.start() and tracker.stop)

• Do the same thing with a neural network to compare:

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"])

• You can observe a new file emissions.csv. Observe all information they give you

• To visualize it with the codecarbon interface, you can use command: `carbonboard --filepath=emissions.csv --port=3000

2- Is this front is eco-friendly or greenwashing?

• Create a new profile into your chrome

• Install chrome extension "GreenIT"

• Go to the website you want to analyze

• into the inspector, go to the GreenIT tab

• clear cache, press CMD-R, check 'Activer l'analyse des bonnes pratiques' and click on 'Lancer l'analyse