Estimate ecological footprint of AI models
Dev Tools
- AIPowerMeter – A library that enables monitoring energy usage of machine learning programs, using RAPL for the CPU and nvidia-smi for the GPU.
- CodeCarbon – Estimates the carbon footprint during the training of machine learning models. (CodeCarbon)
- EnergyMeter – A Python module combining pyRAPL, NVIDIA-SMI, and eBPF to estimate energy consumption of CPU, memory, GPU, and storage on Linux with only three lines of code.
- pyJoules – A Python library that uses hardware measurement tools (Intel RAPL, NVIDIA GPU tools, etc.) to measure device energy consumption. (pyJoules)
- EcoLogits – A Python library that tracks the energy consumption and environmental footprint of using generative AI models through APIs. (EcoLogits)
- Carbontracker – A tool written in Python to predict the carbon footprint of training of models (Paper: https://doi.org/10.48550/arXiv.2007.03051
Online Calculators
- Green Algorithms – Provides an online calculator to estimate the carbon footprint of workloads.
- ML CO2 Impact – Provides an online calculator to estimate the carbon footprint of AI workloads.
- EcoLogits Calculator – Provides an online calculator to estimate the environmental footprint of generative AI models.
- ML.ENERGY Leaderboard – Provides a leaderboard for open source LLMs regarding their energy consumption.
- Green Coding AI – Measures the energy consumption of the inference of open source LLMs. By: Green Coding Solutions
See also: Displaying carbon emissions for your model by Hugging Face
🔗 References
GitHub - navveenb/sustainable-ai: A comprehensive repository to explore resources aimed at reducing the environmental footprint of AI systems by Navveen Balani