Evaluating Environmental Impact of SLMs and Prompt Engineering for Code Generation
arXiv·low signal
As locally-deployed open-source small language models (SLMs) replace cloud-hosted LLMs for AI-assisted coding, the environmental footprint of AI decentralizes. This study measures the energy and carbon impact of different prompting strategies across SLMs for code generation tasks — providing the first empirical sustainability data for the growing cohort of developers running local models for coding.