Eco2AI

Overview

Eco2AI is a Python library for CO2 emission tracking. It monitors the energy consumption of CPU and GPU devices and estimates equivalent carbon emissions using a regional emission coefficient, and the project describes it as applicable to all Python scripts with only a couple of lines added. Every tracked run writes a session description to a local log with project name, experiment description, start time, duration, power consumption, CO2 emissions, CPU and GPU names, operating system and country. The library supports a decorator as well as explicit Tracker start/stop calls, is released under the Apache 2.0 licence, and is developed with the AIRI institute.

Use cases

• Track CO2 emissions of model training runs
• Log CPU and GPU power consumption per experiment
• Compare energy use across experiments
• Estimate emissions with regional carbon intensity

Sources

• Eco2AI repository
• Eco2AI Apache 2.0 licence