Carbon Aware Computing for GenAI Developers, a new course with Google Cloud is live!

I'm delighted to introduce Carbon Aware computing for Genai developers built in partnership with Google Cloud and taught by Nikita Namjushi. Training, fine tuning and even survey models, especially generative AI models, can be compute intensive and energy intensive, but it may not have to be carbon intensive. If you customize where and when you run your jobs in the cloud, factors like training strategy, data center efficiency as well as time and location can have a huge impact on the carbon emissions of a machine learning training job. For example, you can reduce the CO2 footprint of your jobs by identifying and using data centers that are powered by a high amount of renewable or non CO2 emitting energy sources such as solar, wind and hydro. In this short course, you retrieve real time information about where a region's electricity comes from and how much carbon was emitted to produce it by making an API call to the Electricity Maps site and you'll see how you can use such data to inform what cloud region you train a machine learning model in. I'm delighted to introduce the instructor Nikita, who's developer advocate at Google Cloud. Nikita has worked on Carbon Aware computing and sustainability efforts across Google Cloud and google.org for the past five plus years.

Thanks Andrew. In this course you'll learn how machine learning workloads produce greenhouse gas emissions mainly through the electricity that powers the compute. Think the GPUs, TPUs and CPUs, but also from the physical infrastructure at the data centers that power these processors such as cooling infrastructure. You'll explore the energy breakdown and carbon intensity of cities and countries around the world and and maybe even your own city, and calculate what percentage of a particular energy grid's electricity comes from renewable and non renewable sources. You'll also try out one of the simplest and most effective strategies for Carbon Aware ML development training a model in a data center that is powered by a high amount of carbon free energy. Along the way, you'll learn some key terminology about how carbon is quantified and reported according to the internationally recognized Greenhouse Gas Protocol corporate standard. And you'll even look at some real data about carbon emissions produced from some of my recent ML training jobs.

That's great Nikita. For anyone doing development in the cloud, being carbon Aware can help reduce your CO2 footprint I hope you take this course and see what you could do to get your important work in AI done, but perhaps with reduced emissions.

I hope you enjoy the course.