saintslab/carbontracker

Track and predict the energy consumption and carbon footprint of training deep learning models.

What it solves

It provides a way to track and predict the energy consumption and carbon footprint of training deep learning models, helping researchers and developers understand the environmental impact of their AI workloads.

How it works

The tool monitors power usage of CPUs and GPUs during training. It can be used as a command-line wrapper for scripts or embedded directly into Python code. By monitoring a few initial epochs, it can predict the total carbon footprint for the entire training run. It also converts CO2 equivalents (CO2eq) into interpretable metrics, such as the distance a car would travel.

Who it’s for

AI researchers and machine learning engineers who want to quantify and report the environmental cost of training their models.

Highlights

  • Predictive Tracking: Predicts total energy and carbon consumption based on early training epochs.
  • Flexible Integration: Works as a CLI tool or a Python library.
  • Hardware Compatibility: Supports NVIDIA GPUs (via NVML) and Intel CPUs (via RAPL).
  • Reporting: Generates detailed PDF reports and allows for the aggregation of multiple log files into a single estimate.
  • C-level Interpretability: Translates carbon emissions into real-world equivalents like car travel distance.

Related

  • Project
    mlco2/codecarbon

    A lightweight Python library that estimates and tracks carbon emissions from local computing hardware (CPU, GPU, and RAM) to help developers reduce their environmental impact.

  • Dispatch
    Hugging Face Hub Carbon Emissions Tracking

    Hugging Face has introduced tools to track, report, and filter machine learning models based on their CO2 emissions to improve environmental transparency in AI.

  • Project
    ml-energy/zeus

    A library for measuring and optimizing the energy consumption of deep learning workloads across various hardware platforms including NVIDIA, AMD, and Apple Silicon.

  • Project
    atomistic-machine-learning/schnetpack

    A deep learning toolbox for predicting quantum-chemical properties of molecules and materials using atomistic neural networks.

  • Project
    DeepLabCut/DeepLabCut

    DeepLabCut is a markerless pose estimation toolbox that allows researchers to track animals and objects in video without physical markers using deep learning.