针对嵌入式设备微调小模型的高能耗问题，提出节能型GPU动态调压调频方案。
Energy-Efficient GPU DVFS for Fine-Tuning of SLMs on Resource-constrained Embedded Devices
- 基于机器学习选择嵌入式GPU在微调时的最佳电压频率配置。
- 在Jetson AGX Orin上实现平均13.11%节能，最高达26.73%。
- 适合需本地化微调、资源受限的边缘设备部署场景。
在资源受限的嵌入式GPU平台上，动态电压频率调节（DVFS）对于实现小型语言模型（SLM）微调的能效优化至关重要。随着隐私和个性化需求推动本地执行，微调需在多个小批次上进行反复前向-反向传播优化，其耗时与能耗远高于单次推理。为此，我们首先对代表性编码器类SLM（BERT系列）与自回归解码器类SLM（Pythia系列）在GLUE基准上的微调行为进行了分析。此外，提出一种简单但高效的基于机器学习的模型选择方法，用于在资源受限嵌入式平台中选取能量最优的GPU DVFS设置。在NVIDIA Jetson AGX Orin上的实验表明，该方法相比无功耗限制的MAXN Mode 0，平均节能13.11%（最高达26.73%）。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。