Environmental Impacts of Large Language Models

This paper is a research article that evaluates the environmental impacts of Large Language Models (LLMs). It discusses the increasing energy demands associated with LLMs and the necessity to assess their environmental effects as their usage expands globally. The paper outlines the methodologies used in current literature to determine energy consumption and environmental impact during both training and inference stages of LLMs. It presents a toolkit designed for researchers to systematically evaluate these impacts, identifying key factors and variables that influence energy usage and emissions. The article also contextualizes the environmental effects of LLMs by comparing them to previous digital technologies, such as email and blockchain, and emphasizes the importance of understanding the nuances of emissions related to different model architectures and prompting types. The paper aims to contribute to ongoing discussions about sustainability in technology by providing a structured approach to assessing the environmental footprint of LLMs.