{"ID":6023420,"CreatedAt":"2026-07-08T01:00:23.257252134Z","UpdatedAt":"2026-07-10T07:26:08.066495556Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2607.05933","arxiv_id":"2607.05933","title":"Energy-Efficient GPU DVFS for Fine-Tuning of SLMs on Resource-constrained Embedded Devices","abstract":"Dynamic Voltage Frequency Scaling (DVFS) on resource-constrained embedded GPU platforms is essential for energy-efficient small language model (SLM) fine-tuning, as privacy- and personalization-driven adaptation increasingly requires local execution and involves repeated forward-backward optimization over many mini-batches, making it substantially more time- and energy-intensive than single-pass inference. To this end, 1) we first characterize the fine-tuning behavior of representative encoder-only SLMs of BERT variants, and autoregressive decoder-only SLMs of Pythia variants on GLUE benchmarks. In addition to the characterizations, 2) we propose a simple yet effective ML-based model selection that selects energy-optimal GPU DVFS settings on resource-constrained embedded platforms. Our results on NVIDIA Jetson AGX Orin demonstrate average 13.11% energy savings (up to 26.73%) over MAXN Mode 0, which has no explicit power cap.","short_abstract":"Dynamic Voltage Frequency Scaling (DVFS) on resource-constrained embedded GPU platforms is essential for energy-efficient small language model (SLM) fine-tuning, as privacy- and personalization-driven adaptation increasingly requires local execution and involves repeated forward-backward optimization over many mini-bat...","url_abs":"https://arxiv.org/abs/2607.05933","url_pdf":"https://arxiv.org/pdf/2607.05933v1","authors":"[\"Jurn-Gyu Park\",\"Sanzhar Zholdybayev\",\"Aidar Amangeldi\",\"Ademi Zhanuzakova\"]","published":"2026-07-07T07:34:56Z","proceeding":"cs.PF","tasks":"[\"cs.PF\",\"cs.LG\"]","methods":"[\"Language Model\"]","has_code":false}