GAT4ML - Reducing the Environmental Impact of ML-Enabled Systems via Green Architectural Tactics
To combat the increased energy demand of systems using machine learning (ML), especially modern ones based on large language models (LLMs), we take a holistic software architecture perspective to synthesize reusable software design guidance to improve environmental sustainability via green architectural tactics. We will provide a holistic collection of effective green tactics, a tool-supported methodology to guide the continuous synthesis of tactics for researchers, a tool-supported methodology to allow efficient tactic browsing and selection for practitioners, empirical evidence for the effectiveness and trade-offs for tactics, and guidelines for applying and implementing the tactics in real-world industry settings, including prototypical sample implementations.
