Research

Paper

AI LLM March 03, 2026

SEALing the Gap: A Reference Framework for LLM Inference Carbon Estimation via Multi-Benchmark Driven Embodiment

Authors

Priyavanshi Pathania, Rohit Mehra, Vibhu Saujanya Sharma, Vikrant Kaulgud, Tiffani Nevels, Sanjay Podder, Adam P. Burden

Abstract

Large Language Models are rapidly gaining traction in software engineering, yet their growing carbon footprint raises pressing sustainability concerns. While training emissions are substantial, inference quickly surpasses them due to the sheer volume of prompts processed. This shift underscores the urgent need for accurate, prompt-level carbon measurement during inference to enable informed, sustainability-focused decision-making. To address the limitations of existing approaches, in this paper, we outline the guiding principles for a novel reference framework for LLM inference carbon estimation that can guide the design of future tools and provide a systematic foundation for advancing sustainability research in this domain. We also introduce SEAL, an early embodiment of these principles that leverages a multi-benchmark-driven approach for per-prompt carbon estimation. Its initial validation shows promising results, positioning SEAL as a foundation for standardized sustainability assessment across the LLM ecosystem.

Metadata

arXiv ID: 2603.02949
Provider: ARXIV
Primary Category: cs.SE
Published: 2026-03-03
Fetched: 2026-03-04 03:41

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.02949v1</id>\n    <title>SEALing the Gap: A Reference Framework for LLM Inference Carbon Estimation via Multi-Benchmark Driven Embodiment</title>\n    <updated>2026-03-03T12:57:27Z</updated>\n    <link href='https://arxiv.org/abs/2603.02949v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.02949v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Large Language Models are rapidly gaining traction in software engineering, yet their growing carbon footprint raises pressing sustainability concerns. While training emissions are substantial, inference quickly surpasses them due to the sheer volume of prompts processed. This shift underscores the urgent need for accurate, prompt-level carbon measurement during inference to enable informed, sustainability-focused decision-making. To address the limitations of existing approaches, in this paper, we outline the guiding principles for a novel reference framework for LLM inference carbon estimation that can guide the design of future tools and provide a systematic foundation for advancing sustainability research in this domain. We also introduce SEAL, an early embodiment of these principles that leverages a multi-benchmark-driven approach for per-prompt carbon estimation. Its initial validation shows promising results, positioning SEAL as a foundation for standardized sustainability assessment across the LLM ecosystem.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.SE'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <published>2026-03-03T12:57:27Z</published>\n    <arxiv:comment>5 pages. To be published in the proceedings of 48th International Conference on Software Engineering (ICSE '26), April 12-18, 2026, Rio de Janeiro, Brazil (New Ideas and Emerging Results Track)</arxiv:comment>\n    <arxiv:primary_category term='cs.SE'/>\n    <author>\n      <name>Priyavanshi Pathania</name>\n    </author>\n    <author>\n      <name>Rohit Mehra</name>\n    </author>\n    <author>\n      <name>Vibhu Saujanya Sharma</name>\n    </author>\n    <author>\n      <name>Vikrant Kaulgud</name>\n    </author>\n    <author>\n      <name>Tiffani Nevels</name>\n    </author>\n    <author>\n      <name>Sanjay Podder</name>\n    </author>\n    <author>\n      <name>Adam P. Burden</name>\n    </author>\n    <arxiv:doi>10.1145/3786582.3786846</arxiv:doi>\n    <link href='https://doi.org/10.1145/3786582.3786846' rel='related' title='doi'/>\n  </entry>"
}