GreenAccounter: A toolkit for carbon-aware orchestration of deep learning workloads in geo-distributed clouds

Deep learning workloads generate substantial carbon emissions in data centers, largely because training is both time-consuming and energy-intensive. To address this challenge, previous studies have explored either temporal workload shifting or spatial workload migration. Still, these approaches remain limited for long-running workloads, such as Large Language Models (LLMs), because they fail to adapt to continuous fluctuations in regional carbon intensity. In this paper, we introduce GreenAccounter , a toolkit for carbon-aware orchestration of deep learning workloads across multi-cloud environments. It integrates real-time carbon intensity monitoring with checkpoint-based migration, allowing training to continue seamlessly while reducing emissions. A unified dashboard visualizes regional carbon intensity, cumulative emissions, and power consumption, providing operators with a single pane of glass for managing distributed cloud resources. GreenAccounter serves as both (i) a reproducible research platform for carbon-aware scheduling and (ii) a practical operational toolkit for emissions reduction in AI training. As an open-source release, it promotes sustainable, transparent, and data-driven practices for large-scale deep learning. • Implements carbon-aware orchestration for deep learning in multi-cloud environments. • Enables checkpoint-based migration while preserving training states and continuity. • Provides real-time monitoring of carbon intensity and energy metrics. • Offers an open-source platform for reproducible, emissions-aware research.