CodeCarbon API

Warning

This mode use the CodeCarbon API to upload the timeseries of your emissions on a central server.

Thanks to CleverCloud the use of API is free as soon as it remains under certain limits.

Before using it, you need to create on account on the CodeCarbon dashboard

Then login from your terminal:

codecarbon login

It will create an experiment_id for the default project and save it to codecarbon.config

Then you can tell CodeCarbon to monitor your machine:

codecarbon monitor

Or use the API in your code:

from codecarbon import track_emissions

@track_emissions(save_to_api=True)
def train_model():
    # GPU intensive training code  goes here

if __name__ =="__main__":
    train_model()

More options could be specified in @track_emissions or in .codecarbon.config

The CodeCarbon dashboard use CodeCarbon API to get the data

The API do not have a nice web interface to create your own organization and project, you have to use OpenAPI interface for that.

And so on for your team, project and experiment.

You then have to set you experiment id in CodeCarbon, with two options:

In the code:

from codecarbon import track_emissions

@track_emissions(
    measure_power_secs=30,
    api_call_interval=4,
    experiment_id="your experiment id",
    save_to_api=True,
)
def train_model():
    ...

Or in the config file .codecarbon.config:

[codecarbon]
experiment_id = your experiment id
save_to_api = true