检索结果-内蒙古大学图书馆

Automatic parallelism strategy generation with minimalmemory redundancy

Frontiers of Information Technology & Electronic Engineering 2025年第1期26卷 109-118页

作者： Yanqi SHI Peng LIANG Hao ZHENG Linbo QIAO Dongsheng LI National Key Laboratory of Parallel and Distributed Computing National University of Defense TechnologyChangsha 410000China

Large-scale deep learning models are trained distributedly due to memory and computing resource *** existing strategy generation approaches take optimal memory minimization as the *** fill in this gap,we propose a novel algorithm that generates optimal parallelism strategies with the constraint of minimal memory *** propose a novel redundant memory cost model to calculate the memory overhead of each operator in a given parallel *** generate the optimal parallelism strategy,we formulate the parallelism strategy search problem into an integer linear programming problem and use an efficient solver to find minimal-memory intra-operator parallelism ***,the proposed algorithm has been extended and implemented in a multi-dimensional parallel training framework and is characterized by high throughput and minimal memory *** results demonstrate that our approach achieves memory savings of up to 67%compared to the latest Megatron-LM strategies;in contrast,the gap between the throughput of our approach and its counterparts is not large.

关键词： Deep learning Automatic parallelism Minimal memory redundancy

来源：评论

学校读者我要写书评

暂无评论

Training large-scale language models with limited GPU memory:a survey

引用

Frontiers of Information Technology & Electronic Engineering 2025年第3期26卷 309-331页

作者： Yu TANG Linbo QIAO Lujia YIN Peng LIANG Ao SHEN Zhilin YANG Lizhi ZHANG Dongsheng LI National Key Laboratory of Parallel and Distributed Computing College of ComputerNational University of Defense TechnologyChangsha 410073China

Large-scale models have gained significant attention in a wide range of fields,such as computer vision and natural language processing,due to their effectiveness across various ***,a notable hurdle in training these large-scale models is the limited memory capacity of graphics processing units(GPUs).In this paper,we present a comprehensive survey focused on training large-scale models with limited GPU *** exploration commences by scrutinizing the factors that contribute to the consumption of GPU memory during the training process,namely model parameters,model states,and model *** this analysis,we present an in-depth overview of the relevant research work that addresses these aspects ***,the paper concludes by presenting an outlook on the future of memory optimization in training large-scale language models,emphasizing the necessity for continued research and innovation in this *** survey serves as a valuable resource for researchers and practitioners keen on comprehending the challenges and advancements in training large-scale language models with limited GPU memory.

关键词： Training techniques Memory optimization Model parameters Model states Model activations

来源：评论

学校读者我要写书评

暂无评论

An intelligent mesh-smoothing method with graph neural networks

引用

Frontiers of Information Technology & Electronic Engineering 2025年第3期26卷 367-384页

作者： Zhichao WANG Xinhai CHEN Junjun YAN Jie LIU Science and Technology on Parallel and Distributed Processing Laboratory National University of Defense TechnologyChangsha 410073China Laboratory of Digitizing Software for Frontier Equipment National University of Defense TechnologyChangsha 410073China

In computational fluid dynamics(CFD),mesh-smoothing methods are widely used to refine the mesh quality for achieving high-precision numerical ***,optimization-based smoothing is used for high-quality mesh smoothing,but it incurs significant computational *** works have improved its smoothing efficiency by adopting supervised learning to learn smoothing methods from high-quality ***,they pose difficulties in smoothing the mesh nodes with varying degrees and require data augmentation to address the node input sequence ***,the required labeled high-quality meshes further limit the applicability of the proposed *** this paper,we present graph-based smoothing mesh net(GMSNet),a lightweight neural network model for intelligent mesh *** adopts graph neural networks(GNNs)to extract features of the node’s neighbors and outputs the optimal node *** smoothing,we also introduce a fault-tolerance mechanism to prevent GMSNet from generating negative volume *** a lightweight model,GMSNet can effectively smooth mesh nodes with varying degrees and remain unaffected by the order of input data.A novel loss function,MetricLoss,is developed to eliminate the need for high-quality meshes,which provides stable and rapid convergence during *** compare GMSNet with commonly used mesh-smoothing methods on two-dimensional(2D)triangle *** results show that GMSNet achieves outstanding mesh-smoothing performances with 5%of the model parameters compared to the previous model,but offers a speedup of 13.56 times over the optimization-based smoothing.

关键词： Unstructured mesh Mesh smoothing Graph neural network Optimization-based smoothing

来源：评论

学校读者我要写书评

暂无评论

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

引用

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：评论

学校读者我要写书评

暂无评论

FMCC-RT: a scalable and fine-grained all-reduce algorithm for large-scale SMP clusters

引用

Science China(Information Sciences) 2025年第5期68卷 362-379页

作者： Jintao PENG Jie LIU Jianbin FANG Min XIE Yi DAI Zhiquan LAI Bo YANG Chunye GONG Xinjun MAO Guo MAO Jie REN School of Computer Science and Technology National University of Defense Technology Science and Technology on Parallel and Distributed Processing Laboratory National University of Defense Technology Laboratory of Digitizing Software for Frontier Equipment National University of Defense Technology National Supercomputer Center in Tianjin School of Computer Science Shaanxi Normal University

All-reduce is a widely used communication technique for distributed and parallel applications typically implemented using either a tree-based or ring-based scheme. Each of these approaches has its own limitations: tree-based schemes struggle with efficiently exchanging large messages, while ring-based solutions assume constant communication throughput,an unrealistic expectation in modern network communication infrastructures. We present FMCC-RT, an all-reduce approach that combines the advantages of tree-and ring-based implementations while mitigating their drawbacks. FMCC-RT dynamically switches between tree and ring-based implementations depending on the size of the message being processed. It utilizes an analytical model to assess the impact of message sizes on the achieved throughput, enabling the derivation of optimal work partitioning parameters. Furthermore, FMCC-RT is designed with an Open MPI-compatible API, requiring no modification to user code. We evaluated FMCC-RT through micro-benchmarks and real-world application tests. Experimental results show that FMCC-RT outperforms state-of-the-art tree-and ring-based methods, achieving speedups of up to 5.6×.

关键词： all-reduce collective communication MPI scalability

来源：评论

学校读者我要写书评

暂无评论

LSSM-SpMM: A Long-Row Splitting and Short-Row Merging Approach for parallel SpMM on PEZY-SC3s 24th

LSSM-SpMM: A Long-Row Splitting and Short-Row Merging Appro...

引用

24th International Conference on Algorithms and Architectures for parallel processing, ICA3PP 2024

作者： Cao, Ligang Wang, Qinglin Yang, Shun Xia, Rui Guo, Weihao Liu, Jie Laboratory of Digitizing Software for Frontier Equipment National University of Defense Technology Changsha410073 China National Key Laboratory of Parallel and Distributed Computing National University of Defense Technology Changsha410073 China

ISBN: (纸本)9789819615506

Sparse Matrix-Dense Matrix Multiplication (SpMM) is a crucial kernel used in a wide range of fields including machine learning and linear algebra solvers. Thus, enhancing the performance of SpMM is essential. The uneven distribution of non-zeros in sparse matrices and the tight data dependency between sparse and dense matrices make efficiently running SpMM on various hardware platforms challenging. To address these issues, optimisations are tailored according to the characteristics of the different hardware platforms. In this study, we propose a Long Row Split and Short Row Merge (LSSM) approach on the new MIMD computing platform PEZY-SC3s, utilising the standard Compressed Sparse Row (CSR) format to optimise SpMM. Specifically, LSSM divides the rows of the sparse matrix into short rows (rows with a number of non-zeros less than blockSize) and long rows (rows with a number of non-zeros greater than blockSize), applying splitting to the long rows and merging to the short rows to optimize their computations separately. Additionally, based on the hardware features of PEZY-SC3s, we employed the atomic cache for workload scheduling, SIMD instructions to accelerate computation, and Local Memory to reduce data read-write operations, addressing issues of poor data locality, workload imbalance, and vectorisation during SpMM execution. As the first study of SpMM on PEZY-SC3s, compared with BS-SpMM and RoDe-SpMM implemented on PEZY-SC3s, LSSM-SpMM offers up to 26.17× and 13.59× acceleration on the SuiteSparse and deep learning datasets, respectively, with geometric mean speedups of 1.56× and 1.71×. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

DaCP: Accelerating Synchronization-Free SpTRSV via GPU-Friendly Data Communication and parallelism Strategies 20th

DaCP: Accelerating Synchronization-Free SpTRSV via GPU-Frie...

引用

20th IFIP WG 10.3 International Conference on Network and parallel Computing, NPC 2024

作者： Guo, Mingfeng Deng, Liang Dai, Zhe Li, Ruitian Lin, Gaofeng Liu, Jie Computational Aerodynamics Institute China Aerodynamics Research and Development Center Mianyang China Science and Technology on Parallel and Distributed Processing Laboratory National University of Defense Technology Changsha China

ISBN: (纸本)9789819628292

Sparse triangular solve (SpTRSV) is a vital component in various scientific applications, and numerous GPU-based SpTRSV algorithms have been proposed. Synchronization-free SpTRSV is currently the mainstream algorithm on GPU due to its short preprocessing time and outstanding performance. However, we observed that this algorithm still has two performance bottlenecks. Firstly, the thread-level parallel mode can introduce to thread divergence issues within GPU warps during the writing phase. Secondly, the thread-level and warp-level fusion mode may struggles to fully exploit GPU resources due to suboptimal mapping relationships between rows and threads. To address these issues, this paper proposes DaCPSpTRSV, a new synchronization-free algorithm with GPU-friendly data communication and parallelism strategies. Specifically, we first develop a fast-forward thread-level approach, incorporating an efficient global memory access pattern and a light-weight dependency control mechanism, to optimize data communication and alleviate thread divergence. A fine-grained fusion strategy is then proposed to maximize GPU parallelism by adaptively selecting the suitable thread-level or warp-level modes. Moreover, the commonly-used compressed sparse row (CSR) format is employed in our DaCPSpTRSV, enhancing the versatility of our algorithm. We evaluate our approach using 245 matrices from the SuiteSparse Matrix Collection on two NVIDIA GPUs, demonstrating speedup ratios of up to 4.77×, 4.94×, 1.67×, and 1.62× compared to cuSPARSE, Sync-Free, CapelliniSpTRSV, and YuenyeungSpTRSV, respectively. The project is open-sourced at https://***/gmfff12334/DaCP. © IFIP International Federation for Information processing 2025.

关键词： Graphics processing unit

来源：评论

学校读者我要写书评

暂无评论

AFMA-Track: Adaptive Fusion of Motion and Appearance for Robust Multi-object Tracking 27th

AFMA-Track: Adaptive Fusion of Motion and Appearance for ...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Liao, Wei Luo, Lei Zhang, Chunyuan College of Computer Science and Technology National University of Defence Technology Changsha China Science and Technology on Parallel and Distributed Processing Laboratory College of Computer Science and Technology National University of Defense Technology Changsha China

ISBN: (纸本)9783031784439

Motion and appearance cues play a crucial role in Multi-object Tracking (MOT) algorithms for associating objects across consecutive frames. While most MOT methods prioritize accurate motion modeling and distinctive appearance representations, the use of appearance and motion cues is often confined to simplistic association techniques. For instance, fixed weights are commonly employed to combine the intersection-over-union (IoU) matrix and appearance similarity matrix, yielding an association cost matrix. To harness the full potential of motion and appearance cues across diverse scenarios, we propose an innovative approach that dynamically balances motion and appearance cues based on scene and object information during the association process. Furthermore, we introduce a new mechanism for updating appearance representations, effectively mitigating noise introduced by occlusion. Our method demonstrates state-of-the-art performance on the MOT17 and MOT20 test sets. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

Deep Time Series Anomaly Detection with Local Temporal Pattern Learning

Deep Time Series Anomaly Detection with Local Temporal Patte...

引用

2025 IEEE International Conference on Acoustics, Speech, and Signal processing, ICASSP 2025

作者： Li, Yizhou Wang, Yijie Xu, Hongzuo Zhou, Xiaohui National Key Laboratory of Parallel and Distributed Computing College of Computer Science and Technology National University of Defense Technology Changsha410073 China Beijing100091 China

ISBN: (纸本)9798350368741

Self-supervised time series anomaly detection (TSAD) demonstrates remarkable performance improvement by extracting high-level data semantics through proxy tasks. Nonetheless, most existing self-supervised TSAD techniques rely on manual- or neural-based transformations when designing proxy tasks, overlooking the intrinsic temporal patterns of time series. This paper proposes a local temporal pattern learning-based time series anomaly detection (LTPAD). LTPAD first generates sub-sequences. Pairwise sub-sequences naturally manifest proximity relationships along the time axis, and such correlations can be used to construct supervision and train neural networks to facilitate the learning of temporal patterns. Time intervals between two sub-sequences serve as labels for sub-sequence pairs. By classifying these labeled data pairs, our model captures the local temporal patterns of time series, thereby modeling the temporal pattern-aware "normality". Abnormal scores of testing data are acquired by evaluating their conformity to these learned patterns shared in training data. Extensive experiments show that LTPAD significantly outperforms state-of-the-art competitors. © 2025 IEEE.

关键词： Local Temporal Pattern Self-supervised Learning Time Series Anomaly Detection

来源：评论

学校读者我要写书评

暂无评论

Comprehensive Deadlock Prevention for GPU Collective Communication 25

Comprehensive Deadlock Prevention for GPU Collective Communi...

引用

20th European Conference on Computer Systems, EuroSys 2025, co-located 30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOS 2025

作者： Pan, Lichen Liu, Juncheng Fu, Yongquan Yuan, Jinhui Zhang, Rongkai Li, Pengze Xiao, Zhen School of Computer Science Peking University China OneFlow Research China National Key Laboratory of Parallel and Distributed Computing College of Computer Science and Technology National University of Defense Technology China

ISBN: (纸本)9798400711961

distributed deep neural network training necessitates efficient GPU collective communications, which are inherently susceptible to deadlocks. GPU collective deadlocks arise easily in distributed deep learning applications when multiple collectives circularly wait for each other. GPU collective deadlocks pose a significant challenge to the correct functioning and efficiency of distributed deep learning, and no general effective solutions are currently available. Only in specific scenarios, ad-hoc methods, making an application invoke collectives in a consistent order across GPUs, can be used to prevent circular collective dependency and deadlocks. This paper presents DFCCL, a novel GPU collective communication library that provides a comprehensive approach for GPU collective deadlock prevention while maintaining high performance. DFCCL achieves preemption for GPU collectives at the bottom library level, effectively preventing deadlocks even if applications cause circular collective dependency. DFCCL ensures high performance with its execution and scheduling methods for collectives. Experiments show that DFCCL effectively prevents GPU collective deadlocks in various situations. Moreover, extensive evaluations demonstrate that DFCCL delivers performance comparable to or superior to NCCL, the state-of-the-art collective communication library highly optimized for NVIDIA GPUs. © 2025 Copyright held by the owner/author(s).

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

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

请选择保存的检索档案：

请选择收藏分类：

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

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

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：