The rapid advancement and extensive implementation of Large Language Models (LLMs) are milestones in the realm of artificial intelligence. Although Parameter-Efficient Transfer Learning (PETL), a.k.a. Adapter, methods...
详细信息
ISBN:
(纸本)9798350368567;9798350368550
The rapid advancement and extensive implementation of Large Language Models (LLMs) are milestones in the realm of artificial intelligence. Although Parameter-Efficient Transfer Learning (PETL), a.k.a. Adapter, methods have reduced the barrier for fine-tuning and inference on LLMs, it becomes a challenge to efficiently deploy and fine-tuning different adapter models needed for massive AI applications. With the popularity of SoC chips, the computing power of edge devices has improved significantly. To meet the computational resources required by LLM applications and improve quality of service (QoS), we propose Edge-LLM, a server-node collaboration framework for large-scale language model serving, to efficiently utilize edge resources to accelerate LLM fine-tuning and inference in resource-constrained scenarios. In the framework, we implement an adaptive quantization strategy, FM cache mechanism, and value density first (VDF) schedulingalgorithm to reduce GPU overhead and accelerate LLM computation. The experimental results demonstrate that Edge-LLM can significantly improve overall computational speed by a factor of 17, decrease the number of tasks experiencing timeouts by 63%, and reduce GPU overhead by up to 43%.
The uncertainty of electric vehicle (EV) behavior is deemed as a major challenge in online charging scheduling. It may lead to charging congestion to compromise the whole benefits of EV owners and aggregators. Early c...
详细信息
The uncertainty of electric vehicle (EV) behavior is deemed as a major challenge in online charging scheduling. It may lead to charging congestion to compromise the whole benefits of EV owners and aggregators. Early charging is the most efficient way to tackle the dynamic problem. However, it is very challenging for early charging to achieve the adaptive control and minimize electricity bill. In this paper, a price-responsive early charging adaptive control (PRECC) is proposed. The speedup factor is designed as a subtotal of charging demand categorized by electricity price, and it can be determined with only one offline charging optimization through a data-mining method. Due to the strong correlation with electricity price, PRECC can help online scheduling algorithms minimize early charging cost. Since it is not limited by the states of EVs, it can rapidly respond to the variations of base load and electricity price. Besides, with the independent design, it can well match online scheduling algorithms. Computer simulations are made to verify the proposed control. The results show that PRECC can improve the optimality of onlinescheduling by an average of 5.4%. Compared with the traditional early charging strategies, it has obvious advantages in terms of optimality, power capacity utilization, and profitability.
The uncertainty of plug-in electric vehicle (EV) charging behaviour is a crucial factor that not only influences the peak power demand in distribution networks, but also the tariff plans of EV charging service. The un...
详细信息
The uncertainty of plug-in electric vehicle (EV) charging behaviour is a crucial factor that not only influences the peak power demand in distribution networks, but also the tariff plans of EV charging service. The uncertain upstream electricity price considerably complicates the issue regarding how to achieve specific economic goals for distribution network operators (DNOs) while guaranteeing EV users' interest. A rolling horizon scheduling approach based on Genetic algorithm (GA) is proposed in this paper to provide a win-win strategy for both DNOs and EV users. It deals with the online optimal scheduling problem of aggregated EVs in the energy exchange market. The objective of the scheduling strategy is to maximise DNOs' profit margin by charging EVs in the low price time intervals as well as shifting peak charging loads. The operational constraints of EVs' availability and electricity bidding are all considered in the time rolling horizon framework, meaning all this information will be updated, calculated and partially forecasted at each time interval until the end of the day. A case study is carried out with a 33-node distribution network to verify the effectiveness of the proposed scheduling strategy. In detail, specific tariff plans can be determined toward possible values of uncertain market price to satisfy utilities' economic targets. In this way, both individuals and energy providers that participate in the energy market can benefit from the proposed rolling horizon strategy and keep the uncertainty under control.
In this paper we investigate the online video on demand problem, namely having to accept or reject a request for a movie without knowing the future requests. We present online movie-scheduling schemes that implement t...
详细信息
In this paper we investigate the online video on demand problem, namely having to accept or reject a request for a movie without knowing the future requests. We present online movie-scheduling schemes that implement the principles of refusal by choice and delayed notification. A novel way to schedule movies that exploits the knowledge of the distribution of the preference of requests for movies, is shown to have a competitive ratio that outperforms all the previously known schemes in practical situations. In fact, our scheduler has a competitive ratio bounded above by a constant, independent of the number of the users, channels, or movies, in the case that a large fraction of the requests tends to concentrate in a small number of movies. We extend our approach by presenting an "adaptive" randomized scheduler which initially is not aware of the movie popularities but it adapts to it, and achieves the same asymptotic competitive ratio. (C) 2002 Elsevier Science B.V. All rights reserved.
In this paper we investigate the online video on demand problem, namely having to accept or reject a request for a movie without knowing the future requests. We present online movie-scheduling schemes that implement t...
详细信息
暂无评论