Carbontracker

Overview

Carbontracker is an open-source tool for tracking and predicting the energy consumption and carbon footprint of training deep learning models. It wraps a training script from the command line or runs inside Python, monitors GPU and CPU power, and predicts consumption for the remaining epochs. It can log results to files and generate PDF reports, and supports Slurm and container environments. It is described in the paper by Anthony, Kanding and Selvan (2020).

Use cases

• Track GPU and CPU energy during training
• Predict total run consumption
• Log carbon data for reports
• Monitor cluster jobs

Sources

• Carbontracker repository
• Carbontracker MIT licence