Carbon-Taxed Transformers: Efficient, Accurate, and Sustainable LLMs
炭素課金トランスフォーマー: 効率的で正確かつ持続可能なLLM (AI 翻訳)
Ajmain Inqiad Alam, Palash Ranjan Roy, Chanchal K. Roy, Banani Roy, Kevin A. Schneider
🤖 gxceed AI 要約
日本語
LLMの計算・環境コスト削減のため、経済的炭素課税に着想を得た圧縮パイプライン「Carbon-Taxed Transformers (CTT)」を提案。ニューラルアーキテクチャ探索、構造化プルーニング、量子化、知識蒸留を統合。コードクローン検出、要約、生成タスクで評価し、最大49倍のメモリ削減、10倍の高速化、81%のCO2削減を達成。精度は98%以上を維持。
English
We propose Carbon-Taxed Transformers (CTT), a compression pipeline inspired by economic carbon taxation that integrates Neural Architecture Search, structured pruning, quantization, and knowledge distillation. Evaluated on code clone detection, summarization, and generation, CTT achieves up to 49x memory reduction, 10x speedup, and 81% reduction in CO2 emissions while retaining 98% accuracy on clone detection and >89% on generation tasks.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
日本のGX政策では、AIのエネルギー消費増加が課題となっている。本論文は、LLMの炭素排出を大幅に削減する実用的手法を提供し、日本企業のグリーンIT戦略に直接活用可能。SSBJの開示にもつながるエネルギー効率改善策として注目される。
In the global GX context
Globally, the environmental cost of large AI models is under scrutiny. This paper offers a principled, generalizable pipeline for reducing LLM carbon footprint, aligning with emerging disclosure standards (e.g., TCFD, ISSB) that include Scope 2 and 3 emissions from compute. It provides a replicable method for any organization deploying LLMs to demonstrate sustainability improvements.
👥 読者別の含意
🔬研究者:A novel compression pipeline that integrates multiple techniques with clear environmental impact metrics. Useful for AI efficiency research.
🏢実務担当者:Provides a ready-to-implement method to reduce energy and cost of running LLMs, with measured CO2 reduction.
🏛政策担当者:Demonstrates a technology pathway for reducing AI's carbon footprint, informing potential energy efficiency standards for AI systems.
📄 Abstract（原文）
The growing adoption of Large Language Models in software engineering has introduced unsustainable computational and environmental costs. We propose Carbon-Taxed Transformers (CTT), a compression pipeline inspired by economic carbon taxation, integrating Neural Architecture Search, structured pruning, quantization, and knowledge distillation in a principled sequence. CTT generalizes across encoder-only, encoder-decoder, and decoder-only architectures, and is evaluated on code clone detection, summarization, and generation. Results show up to 49× memory reduction, 10× speedup, and an 81% reduction in CO2 emissions, while retaining 98% accuracy in clone detection, 89% in summarization, and up to 91% in generation. Ablation studies confirm both the pipeline ordering and each component are essential.
🔗 Provenance — このレコードを発見したソース
- openalex https://doi.org/10.1145/3803437.3807387first seen 2026-07-26 05:30:03
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。