Wenxiang Lin

Wenxiang Lin

Auto-generated profile, based on publicly available information

Wenxiang Lin

Harbin Institute of Technology,Shenzhen, Shenzhen, Guangdong, China

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
5 August 2022 in Sensors
Verified
Journal Article
Open access
PDF file
JLJiaxin Li
Jiaxin Li
YDYan Ding
Yan Ding
Hua-Liang WeiHua-Liang Wei
Hua-Liang Wei
Yutong ZhangYutong Zhang
Yutong Zhang
WLWenxiang Lin
Wenxiang Lin
Joint detection and embedding (JDE) methods usually fuse the target motion information and appearance information as the data association matrix, which could fail when the target is briefly lost or blocked in multi-object tracking (MOT). In this paper, we aim to solve this problem by proposing a novel association matrix, the Embedding and GioU (EG) matrix, which combines the embedding cosine distance and GioU distance of objects. To improve the performance of data association, we develop a simple, effective, bottom-up fusion tracker for re-identity features, named SimpleTrack, and propose a new tracking strategy which can mitigate the loss of detection targets. To show the effectiveness of the proposed method, experiments are carried out using five different state-of-the-art JDE-based methods. The results show that by simply replacing the original association matrix with our EG matrix, we can achieve significant improvements in IDF1, HOTA and IDsw metrics, and increase the tracking speed of these methods by around 20%. In addition, our SimpleTrack has the best data association capability among the JDE-based methods, e.g., 61.6 HOTA and 76.3 IDF1, on the test set of MOT17 with 23 FPS running speed on a single GTX2080Ti GPU.

30 March 2025
Verified
Conference Paper
Open access
XPXinglin Pan
Xinglin Pan
WLWenxiang Lin
Wenxiang Lin
LZLin Zhang
Lin Zhang
SSShaohuai Shi
Shaohuai Shi
ZTZhenheng Tang
Zhenheng Tang
XCXiaowen Chu
Xiaowen Chu
Recent large language models (LLMs) have tended to leverage sparsity to reduce computations, employing the sparsely activated mixture-of-experts (MoE) technique. MoE introduces four modules, including token routing, token communication, expert computation, and expert parallelism, that impact model quality and training efficiency. To enable ver- satile usage of MoE models, we introduce FSMoE, a flexible training system optimizing task scheduling with three novel techniques: 1) Unified abstraction and online profiling of MoE modules for task scheduling across various MoE implementations. 2) Co-scheduling intra-node and inter-node communications with computations to minimize communication overheads. 3) To support near-optimal task scheduling, we design an adaptive gradient partitioning method for gradient aggregation and a schedule to adaptively pipeline communications and computations. We conduct extensive experiments with configured MoE layers and real-world MoE models on two GPU clusters. Experimental results show that 1) our FSMoE supports four popular types of MoE routing functions and is more efficient than existing implementations (with up to a 1.42× speedup), and 2) FSMoE outperforms the state-of-the-art MoE training systems (DeepSpeed-MoE and Tutel) by 1.18×-1.22× on 1458 MoE layers and 1.19×-3.01× on real-world MoE models based on GPT-2 and Mixtral using a popular routing function. In this work, we present a flexible training system named FSMoE to optimize task scheduling. To achieve this goal: 1) we design unified abstraction and online profiling of MoE modules across various MoE implementations, 2) we co-schedule intra-node and inter-node communications with computations to minimize communication overhead, and 3) we design an adaptive gradient partitioning method for gradient aggregation and a schedule to adaptively pipeline communications and computations. Experimental results on two clusters up to 48 GPUs show that our FSMoE outperforms the state-of-the-art MoE training systems (DeepSpeed-MoE and Tutel) with speedups of 1.18x-1.22x on 1458 customized MoE layers and 1.19x-3.01x on real-world MoE models based on GPT-2 and Mixtral.

11 August 2026
Verified
Conference Paper
Open access
WLWenxiang Lin
Wenxiang Lin
XPXinglin Pan
Xinglin Pan
RFRuibo Fan
Ruibo Fan
SSShaohuai Shi
Shaohuai Shi
XCXiaowen Chu
Xiaowen Chu
Communication has emerged as a critical bottleneck in the distributed training of large language models (LLMs). While numerous approaches have been proposed to reduce communication overhead, the potential of lossless compression has remained largely underexplored since compression and decompression typically consume larger overheads than the benefits of reduced communication traffic. We observe that the communication data, including activations, gradients and parameters, during training often follows a near-Gaussian distribution, which is a key feature for data compression. Thus, we introduce ZipCCL, a lossless compressed communication library of collectives for LLM training. ZipCCL is equipped with our novel techniques: (1) theoretically grounded exponent coding that exploits the Gaussian distribution of LLM tensors to accelerate compression without expensive online statistics, (2) GPU-optimized compression and decompression kernels that carefully design memory access patterns and pipeline using communication-aware data layout, and (3) adaptive communication strategies that dynamically switch collective operations based on workload patterns and system characteristics. Evaluated on a 64-GPU cluster using both mixture-of-experts and dense transformer models, ZipCCL reduces communication time by up to 1.35X and achieves end-to-end training speedups of up to 1.18X without any impact on model quality.


11 June 2026 in IEEE Transactions on Computers
Verified
Journal Article
Closed access
NWNe Wang
Ne Wang
WLWenxiang Lin
Wenxiang Lin
LZLin Zhang
Lin Zhang
SSShaohuai Shi
Shaohuai Shi
Ruiting ZhouRuiting Zhou
Ruiting Zhou
BLBo Li
Bo Li
Sparsely activated Mixture-of-Experts (MoE) has emerged as a key technique to expand the size of Transformer-based large language models (LLMs) while maintaining low computational costs. However, MoE layers require to route the input data to distributed devices, incurring significant communication latency. Existing studies have primarily focused on alleviating this problem by overlapping computation and communication tasks within a single MoE layer, which fails to achieve sufficient overlap and results in limited performance gains. In this work, we introduce an orthogonal partitioning dimension from existing task-parallel methods by leveraging the autoregressive nature of causal Transformer-based LLMs, i.e. partitioning tasks along the sequence dimension. This provides more flexible and efficient overlaps among tasks from both non-MoE and MoE layers. To this end, we propose an efficient MoE training approach, SP-MoE, with two innovative designs. 1) It incorporates non- MoE layers into the overlapping with not only the current MoE layer but also the preceding MoE layer, thereby facilitating more efficient training; 2) It identifies the optimal combination of pipeline degrees for non-MoE and MoE layers and devises the best scheduling plans for load-imbalanced non-MoE and uniform MoE layers to achieve the goal of minimizing the total training latency. Furthermore, we extend our algorithm to incorporate pipeline parallelism (PP) for training larger MoE models, which introduces peer-to-peer (P2P) communications between adjacent PP stages. To mitigate this problem, we propose S3P-MoE with two optimizations: 1) it partitions P2P communication into chunks and each P2P chunk is transmitted immediately to be overlapped with expert computation; 2) it allows the subsequent PP stage to commence upon receiving the corresponding P2P chunk. Moreover, we refine our algorithm to jointly optimize pipeline degrees for both MoE and non-MoE layers, as well as P2P communication, to minimize overall training latency. Extensive experiments demonstrate that SP-MoE and S3P-MoE effectively identify optimal pipeline degrees, reducing training time by up to 34:3% and 30:61%, respectively, compared to stateof- the-art methods.

13 August 2025 in arXiv
Verified
Preprint
Open access
PDF file
WLWenxiang Lin
Wenxiang Lin
XPXinglin Pan
Xinglin Pan
LZLin Zhang
Lin Zhang
SSShaohuai Shi
Shaohuai Shi
XWXuan Wang
Xuan Wang
XCXiaowen Chu
Xiaowen Chu
The sparsely activated mixture-of-experts (MoE) transformer has become a common architecture for large language models (LLMs) due to its sparsity, which requires fewer computational demands while easily scaling the model size. In MoE models, each MoE layer requires to dynamically choose tokens to activate particular experts for computation while the activated experts may not be located in the same device or GPU as the token. However, this leads to substantial communication and load imbalances across all GPUs, which obstructs the scalability of distributed systems within a GPU cluster. To this end, we introduce HierMoE to accelerate the training of MoE models by two topology-aware techniques: 1) token deduplication to reduce the communication traffic, and 2) expert swap to balance the workloads among all GPUs. To enable the above two proposed approaches to be more general, we build theoretical models aimed at achieving the best token duplication and expert swap strategy under different model configurations and hardware environments. We implement our prototype HierMoE system atop Megatron-LM and conduct experiments on a 32-GPU cluster with DeepSeek-V3 and Qwen3-30B-A3B models. Experimental results show that our HierMoE achieves $1.55\times$ to $3.32\times$ faster communication and delivers $1.18\times$ to $1.27\times$ faster end-to-end training compared to state-of-the-art MoE training systems, Tutel-2DH, SmartMoE, and Megatron-LM.