Characterizing the Carbon Impact of LLM Inference

Edwin Lim, Zhiyang Pan, Yang Zhou

This is a fun class project I completed with my friends that quantifies the carbon impact of Large Language Model (LLM) inference. As it turns out, these models are quite energy-intensive, and can contribute significantly to global carbon emissions, with the severity depending on the carbon intensity of energy within the region. As LLMs become more popular and integrated into everyday life, their overall contribution to carbon emissions from the tech industry should be monitored closely and mitigated. I think we should build more nuclear power plants!