OpenCarbonEval: How much $CO_2$ will your large model exhale in training process?

Zhaojian Yu, Yinghao Wu, Zhuotao Deng, Xinchun Yu, Yansong Tang, Xiao-Ping Zhang

Keywords: Large-scale model, Carbon footprint, Sustainable AI

Abstract:

Data, model and hardware are crucial components in the development of large scale machine learning models. The training of such models necessitates substantial computational resources, energy consumption, and raw materials, resulting in significant environmental implications. However, the environmental impact of these models has been largely overlooked due to a lack of assessment and analysis of their carbon footprint. In this paper, we present OpenCarbonEval, a carbon emission estimation framework to quantify the environmental implications of large scale machine learning models given their total training computations and hardware configurations.
In OpenCarbonEval, we conducted a comprehensive dynamic analysis of the interrelationships among data, models, and hardware throughout the model training process, aiming to forecast the carbon emission of large scale models more accurately. We validated our approach on real-world dataset, and experimental results demonstrate that OpenCarbonEval can predict energy costs and carbon emissions more accurately than previous methods. Furthermore, it can be seamlessly applied to various machine learning tasks without a precision decline. By quantifying the environmental impact of large-scale models, OpenCarbonEval promotes sustainable AI development and deployment, contributing to a more environmentally responsible future for the AI community.

Primary Area: alignment, fairness, safety, privacy, and societal considerations

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 14115