Massively pre-trained transformer models are computationally expensive to fine-tune, slow for inference, and have large storage requirements. Recent approaches tackle these shortcomings by training smaller models, dynamically reducing the model size, and by training light-weight adapters. In this paper, we propose AdapterDrop, removing adapters from lower transformer layers during training and inference, which incorporates concepts from all three directions. We show that Adap-terDrop can dynamically reduce the computational overhead when performing inference over multiple tasks simultaneously, with minimal decrease in task performances. We further prune adapters from AdapterFusion, which improves the inference efficiency while maintaining the task performances entirely. 1 We experiment with newer and older GPUs, Nvidia V100 and Titan X, respectively. See Appendix A.1 for details. 2 We include detailed plots in Appendix G.1.
Almost all power electronic converters require magnetic devices and some properties of the converters can be controlled with magnetics. New magnetic materials enable implementation of magnetic control of converters also in high switching frequency applications. The presented paper gives an overview about controllability of magnetic devices in theory and practice, simulation approaches and examples. The focus is on electrically controllable magnetic devices. This means that an additional control winding is wound on or integrated into the magnetic core of the device. This additional winding is used to control the inductive behavior of the device by injecting a current. The different basic control approaches of magnetic devices are explained theoretically and demonstrated experimentally by laboratory prototypes. Several controllable magnetic devices were tested in order to evaluate the applicability of the different control methods. The control techniques were demonstrated also with different core materials to address different applications. Especially ferromagnetic alloys and ferrite materials are in focus of the investigation. Moreover, first approaches for the modelling and simulation of controllable magnetic devices are given in the paper.
This paper presents a 4 kW GaN-based isolated bidirectional DC/DC converter suitable for on-board auxiliary power supply systems, interconnecting the HV batteries with the LV on-board power supply system. To enable the trend towards higher levels of electrification, a second LV level of 48 V is introduced to supply the high-power consumers, relieving the 12 V grid on the one hand and/or avoiding using additional costly isolated converters connected to the HV batteries on the other. The built converter prototype aims to demonstrate higher efficiency, compactness and lower filter requirements of the proposed circuit.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.