检索结果-南通市图书馆

Optimizing Fine-Tuning in Quantized Language Models:An In-Depth Analysis of Key Variables

Computers, Materials & Continua 2025年第1期82卷 307-325页

作者： Ao Shen Zhiquan Lai dongsheng li Xiaoyu Hu National Key Laboratory of Parallel and Distributed Computing National University of Defense TechnologyChangsha410073China Strategic Assessments and Consultation Institute Academy of Military ScienceBeijing100091China

Large-scale Language Models(LLMs)have achieved significant breakthroughs in Natural Language Processing(NLP),driven by the pre-training and fine-tuning *** this approach allows models to specialize in specific tasks with reduced training costs,the substantial memory requirements during fine-tuning present a barrier to broader ***-Efficient Fine-Tuning(PEFT)techniques,such as Low-Rank Adaptation(LoRA),and parameter quantization methods have emerged as solutions to address these challenges by optimizing memory usage and computational *** these,QLoRA,which combines PEFT and quantization,has demonstrated notable success in reducing memory footprints during fine-tuning,prompting the development of various QLoRA *** these advancements,the quantitative impact of key variables on the fine-tuning performance of quantized LLMs remains *** study presents a comprehensive analysis of these key variables,focusing on their influence across different layer types and depths within LLM *** investigation uncovers several critical findings:(1)Larger layers,such as MLP layers,can maintain performance despite reductions in adapter rank,while smaller layers,like self-attention layers,aremore sensitive to such changes;(2)The effectiveness of balancing factors depends more on specific values rather than layer type or depth;(3)In quantization-aware fine-tuning,larger layers can effectively utilize smaller adapters,whereas smaller layers struggle to do *** insights suggest that layer type is a more significant determinant of fine-tuning success than layer depth when optimizing quantized ***,for the same discount of trainable parameters,reducing the trainable parameters in a larger layer is more effective in preserving fine-tuning accuracy than in a smaller *** study provides valuable guidance for more efficient fine-tuning strategies and opens avenues for further research into optimizing LLM

关键词： Large-scale Language Model Parameter-Efficient Fine-Tuning parameter quantization key variable trainable parameters experimental analysis

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Improved Heuristic Job Scheduling Method to Enhance Throughput for Big Data Analytics

引用

Tsinghua Science and Technology 2022年第2期27卷 344-357页

作者： Zhiyao Hu dongsheng li the College of Computer National University of Defense TechnologyChangsha 410073China

Data-parallel computing platforms,such as Hadoop and Spark,are deployed in computing clusters for big data *** is a general tendency that multiple users share the same computing *** schedule of multiple jobs becomes a serious *** a long period in the past,the Shortest-Job-First(SJF)method has been considered as the optimal solution to minimize the average job completion ***,the SJF method leads to a low system throughput in the case where a small number of short jobs consume a large amount of *** factor prolongs the average job completion *** propose an improved heuristic job scheduling method,called the Densest-Job-Set-First(DJSF)*** DJSF method schedules jobs by maximizing the number of completed jobs per unit time,aiming to decrease the average Job Completion Time(JCT)and improve the system *** perform extensive simulations based on Google cluster *** with the SJF method,the DJSF method decreases the average JCT by 23.19% and enhances the system throughput by 42.19%.Compared with Tetris,the job packing method improves the job completion efficiency by 55.4%,so that the computing platforms complete more jobs in a short time span.

关键词： big data job scheduling job throughput job completion time job completion efficiency

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

A radial symmetry and liouville theorem for systems involving fractional Laplacian

引用

Frontiers of Mathematics in China 2017年第2期12卷 389-402页

作者： dongsheng li Zhenjie li School of Mathematics and Statistics Xi'an Jiaotong University Xi'an 710049 China

We investigate the nonnegative solutions of the system involving the fractional Laplacian：{（-△）^αui（x）=fi（u）,x∈R^n,i=1,2,…,m, u（x）=（u1（x）,u2（x）,……,um（x））,where 0 〈 α 〈 1, n 〉 2, fi（u）, 1 4 ≤ 4 ≤m, are real-valued nonnegative functions of homogeneous degree Pi ≥0 and nondecreasing with respect to the independent variables ul, u2,..., urn. By the method of moving planes, we show that under the above conditions, all the positive solutions are radially symmetric and monotone decreasing about some point x0 if Pi = （n ＋ 2α）/（n- 2α） for each 1 ≤ i ≤ m; and the only nonnegative solution of this system is u ≡ 0 if 1〈pi〈（n＋2α）/（n-2α） for all 1≤i≤m.

关键词： Fractional Laplacian method of moving planes Kelvin transform,liouville theorem

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Effect of Carbon Nanotube on the Oscillating Wear Behaviour of Metal-PTFE Multilayer Composites

引用

Journal of Wuhan University of Technology(Materials Science) 2018年第5期33卷 1271-1274页

作者： ZOU Aili li dongsheng School of Mechanical Engineering ＆ Automation Beihang UniversityBeijing 100191China School of Aerospace Engineering Shenyang A erospace UniversityShenyang 110136China

Two kinds of metal-PTFE multilayer composites, which were composed of a steel backing, a middle layer of sintered porous bronze and a surface layer of polytetrafluoroethylene（PTFE） filled by carbon nanotubes（CNTs） or not, were prepared. The wear properties of metal-PTFE multilayer composites oscillating against 45 carbon steel under dry condition were evaluated on an oscillating wear tester, and the effect of CNTs on wear behaviour of metal-PTFE multilayer composites was studied. The results showed that the worn surface of metal-PTFE multilayer composites was characterized by adhesive wear, abrasive wear and fatigue wear. The CNTs greatly increased the adhesion strength of PTFE in the metal-PTFE composites and thereby greatly reduced puck, ploughing, and fatigue failure of PTFE during wearing. The PTFE filled with CNTs prevented direct contact between the mating surfaces and served as fine self-lubricating film, in which the oscillating wear mechanism of the composites was changed to a slightly adhesive wear. Therefore, the CNTs significantly decreased the weight loss and obviously increased the wear resistance of metal-PTFE multilayer composites.

关键词： carbon nanotubes Metal-PTFE composites oscillating wear self-lubricating

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Balance Resource Allocation for Spark Jobs Based on Prediction of the Optimal Resource

引用

Tsinghua Science and Technology 2020年第4期25卷 487-497页

作者： Zhiyao Hu dongsheng li Deke Guo the College of Computer National University of Defense TechnologyChangsha 410073China the College of System Engineering National University of Defense TechnologyChangsha 410073China

Apache Spark provides a well-known Map Reduce computing framework,aiming to fast-process big data analytics in data-parallel *** this platform,large input data are divided into data *** data partition is processed by multiple computation tasks *** of these computation tasks are transferred among multiple computers via the ***,such a distributed computing framework suffers from system overheads,inevitably caused by communication and disk I/O *** overheads take up a large proportion of the Job Completion Time(JCT).We observed that excessive computational resources incurs considerable system overheads,prolonging the *** over-allocation of individual jobs not only prolongs their own JCTs,but also likely makes other jobs suffer from ***,the average JCT is suboptimal,*** address this problem,we propose a prediction model to estimate the changing JCT of a single Spark *** the support of the prediction method,we designed a heuristic algorithm to balance the resource allocation of multiple Spark jobs,aiming to minimize the average JCT in multiple-job *** implemented the prediction model and resource allocation method in Re B,a Resource-Balancer based on Apache *** results showed that Re B significantly outperformed the traditional max-min fairness and shortest-job-optimal *** average JCT was decreased by around 10%–30%compared to the existing solutions.

关键词： Spark jobs resource over-allocation performance prediction

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Review on residual stress and its effects on manufacturing of aluminium alloy structural panels with typical multi-processes

引用

Chinese Journal of Aeronautics 2023年第5期36卷 96-124页

作者： Yong li Wanni GAN Wenbin ZHOU dongsheng li School of Mechanical Engineering and Automation Beihang UniversityBeijing 100191China Shenzhen Institute of Beihang University Shenzhen 518057China Department of Mechanical Engineering Imperial College LondonLondon SW72AZUK

In the aerospace industry,integrated aluminium alloy plates and stiffened panels with high accuracy and performance attract significant *** manufacture these panels as integrity with high accuracy,multiple processes need to be utilised,such as machining,welding and *** the whole manufacturing chain,residual stresses can be generated and redistributed in the components among different *** residual stress would significantly affect the shapes and properties of the final ***,these great effects are not well considered in the design and manufacturing *** paper aims to draw a general understanding of the residual stress generated in the pre-manufacturing processes and its effects on subsequent manufacturing *** mechanisms and distributions of residual stresses generated in typical premanufacturing processes of structural panels,including machining,welding and additive manufacturing(AM),are firstly *** detailed effects of generated residual stresses on distortion and application properties in subsequent manufacturing processes are then *** addition,current methods developed for the investigation of residual stress effect in multi-processes manufacturing are critically reviewed,including experimental,analytical,finite element(FE)and machine learning ***,the future development trend of methods for residual stress consideration and control in the design of manufacturing processes is summarised,providing comprehensive guidance to achieve the high accurate manufacturing of aluminium alloy structural components.

关键词： Residual stress Stiffened panel Aluminium alloy Multi-processes manufacturing Machining Welding Additive manufacturing

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Large-scale virtual machines provisioning in clouds： challenges and approaches

引用

Frontiers of Computer Science 2016年第1期10卷 2-18页

作者： Zhaoning ZHANG dongsheng li Kui WU National Laboratory for Parallel and Distributed Processing School of Computer Science National University of Defense Technology Changsha 410073 China Computer Science Department University of Victoria Victoria V8W 2Y2 Canada

The scale of global data center market has been explosive in recent years. As the market grows, the demand for fast provisioning of the virtual resources to support elas- tic, manageable, and economical computing over the cloud becomes high. Fast provisioning of large-scale virtual ma- chines （VMs）, in particular, is critical to guarantee quality of service （QoS）. In this paper, we systematically review the existing VM provisioning schemes and classify them in three main categories. We discuss the features and research status of each category, and introduce two recent solutions, VMThunder and VMThunder＋, both of which can provision hundreds of VMs in seconds.

关键词： cloud computing IaaS large scale virtual ma-chine provisioning

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Chitosan foam reinforced by SiC whisker for building insulation with high mechanical strength and vapor permeability

引用

Science China(Technological Sciences) 2022年第12期65卷 2874-2882页

作者： li dongsheng GAO Gang HUANG Congliang School of Low-carbon Energy and Power Engineering China University of Mining and TechnologyXuzhou 221116China

Porous foam based on renewable materials has attracted extensive attention in green energy conservation and sustainable ***,there is still a requisite for biomass-based porous foam that could meet the demand for excellent mechanical and high thermal insulation performances for building ***,we demonstrated a facile strategy to prepare a porous foam made from a chitosan matrix reinforced by SiC whisker,which shows good performance in building insulation and mechanical *** prepared porous foam has a low density(20.1–54.4 kg m^(-3))and high porosity(>97.0%).The density of the chitosan-SiC porous foam can be controlled by varying the suspension solid and SiC whisker content in the preparation ***,the influences of SiC whisker content on the mechanical properties and thermal conductivity of the porous foams were also *** the content of SiC whisker is 30 wt%,the prepared porous foam has the highest compression modulus of about 89.8 kPa and also low thermal conductivity of 0.0354 W m^(-1)K^(-1).Moreover,the prepared porous foam shows excellent water vapor permeability with a vapor resistance factor of 2.94,which is beneficial for moisture transfer in *** work provides a facile way to manufacture biomass-based porous foam,which is potential for energy saving in buildings.

关键词： porous foam biomass-based material thermal insulation mechanical property vapor permeability

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Efficient Multi-Tenant Virtual Machine Allocation in Cloud Data Centers

引用

Tsinghua Science and Technology 2015年第1期20卷 81-89页

作者： Jiaxin li dongsheng li Yuming Ye Xicheng Lu the National Key Laboratory of Parallel and Distributed Processing (PDL) College of Computer National University of Defense Technology

Virtual Machine（VM） allocation for multiple tenants is an important and challenging problem to provide efficient infrastructure services in cloud data centers. Tenants run applications on their allocated VMs, and the network distance between a tenant＇s VMs may considerably impact the tenant＇s Quality of Service（Qo S）. In this study, we define and formulate the multi-tenant VM allocation problem in cloud data centers, considering the VM requirements of different tenants, and introducing the allocation goal of minimizing the sum of the VMs＇ network diameters of all tenants. Then, we propose a Layered Progressive resource allocation algorithm for multi-tenant cloud data centers based on the Multiple Knapsack Problem（LP-MKP）. The LP-MKP algorithm uses a multi-stage layered progressive method for multi-tenant VM allocation and efficiently handles unprocessed tenants at each stage. This reduces resource fragmentation in cloud data centers, decreases the differences in the Qo S among tenants, and improves tenants＇ overall Qo S in cloud data centers. We perform experiments to evaluate the LP-MKP algorithm and demonstrate that it can provide significant gains over other allocation algorithms.

关键词： virtual machine allocation cloud data center multiple tenants multiple knapsack problem

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Effects of Germanium on Movement of Dislocations in p-Type Czochralski Silicon

引用

Journal of Rare Earths 2006年第z1期24卷 83-86页

作者： li dongsheng Zhao Yiying Yang Deren State Key Laboratory of Silicon Materials Zhejiang University Hangzhou 310027 China State Key Laboratory of Silicon Materials Zhejiang University Hangzhou 310027 China State Key Laboratory of Silicon Materials Zhejiang University Hangzhou 310027 China

By indentation at room temperature followed by annealing at high temperatures, the pinning effect of germanium on dislocations in germanium-doped Czochralski silicon was investigated. Experimental results show that the dislocations in germanium-doped Czochralski silicon move shorter and slower than those in Czochralski silicon undoping with germanium when the concentration of germanium is over 1×1018 cm-3. The retarding velocity of dislocations is contributed to the dislocations pinning effect of the strain field introduced by the high concentration germanium, and the Ge4B cluster and the oxygen precipitation those are preferred to form at higher concentration germanium.

关键词： Czochralski silicon germanium-doped dislocation

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

欢迎您,

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

请选择保存的检索档案：

请选择收藏分类：

通借通还

欢迎您,

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：