与电池共舞：在移动设备上实现具有运行时间可重配置性的Transformer
Dancing along Battery: Enabling Transformer with Run-time Reconfigurability on Mobile Devices
摘要：
提出了一种基于剪枝的AutoML运行时间可重构框架RT 3。这使得基于转换器的大型自然语言处理（NLP）模型能够在资源受限的移动设备上高效地执行并重新配置（即，用于动态硬件条件的切换模型）。这种可重构性是电池供电的移动设备节省能量的关键，其广泛使用动态电压和频率缩放（DVFS）技术进行硬件重构以延长电池寿命。在这项工作中，我们创造性地探索了一种混合块结构修剪（BP）和模式修剪（PP）的变压器为基础的模型，并首次尝试结合硬件和软件的重新配置，以最大限度地节省能源的电池供电的移动设备。具体而言，RT 3集成了两级优化：首先，它利用高效的BP作为资源受限的移动设备的第一步压缩;然后，RT 3基于第一级优化来启发性地生成收缩的搜索空间，并通过强化学习来搜索PP的具有不同稀疏度的多个模式集，以支持轻量级软件重配置，其对应于DVFS的可用频率水平（即，硬件重新配置）。在运行时间，RT 3可以在45 ms内切换轻量级模式集，以保证在不同频率水平下所需的实时约束。结果进一步表明，RT 3可以延长电池寿命超过4倍的改进，Transformer的准确性损失小于1%，DistilBERT的得分降低1.5%。
英文摘要：
A pruning-based AutoML framework for run-time reconfigurability, namely RT3, is proposed in this work. This enables Transformer-based large Natural Language Processing (NLP) models to be efficiently executed on resource-constrained mobile devices and reconfigured (i.e., switching models for dynamic hardware conditions) at run-time. Such reconfigurability is the key to save energy for battery-powered mobile devices, which widely use dynamic voltage and frequency scaling (DVFS) technique for hardware reconfiguration to prolong battery life. In this work, we creatively explore a hybrid block-structured pruning (BP) and pattern pruning (PP) for Transformer-based models and first attempt to combine hardware and software reconfiguration to maximally save energy for battery-powered mobile devices. Specifically, RT3 integrates two-level optimizations: First, it utilizes an efficient BP as the first-step compression for resource-constrained mobile devices; then, RT3 heuristically generates a shrunken search space based on the first level optimization and searches multiple pattern sets with diverse sparsity for PP via reinforcement learning to support lightweight software reconfiguration, which corresponds to available frequency levels of DVFS (i.e., hardware reconfiguration). At run-time, RT3 can switch the lightweight pattern sets within 45ms to guarantee the required real-time constraint at different frequency levels. Results further show that RT3 can prolong battery life over 4x improvement with less than 1% accuracy loss for Transformer and 1.5% score decrease for DistilBERT.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
