Zhong Lin Wang

Jinyuan Wang, Hai Zhao, Zhong Wang et al.

In recent years, great advances in pre-trained language models (PLMs) have sparked considerable research focus and achieved promising performance on the approach of dense passage retrieval, which aims at retrieving relative passages from massive corpus with given questions. However, most of existing datasets mainly benchmark the models with factoid queries of general commonsense, while specialised fields such as finance and economics remain unexplored due to the deficiency of large-scale and high-quality datasets with expert annotations. In this work, we propose a new task, policy retrieval, by introducing the Chinese Stock Policy Retrieval Dataset (CSPRD), which provides 700+ prospectus passages labeled by experienced experts with relevant articles from 10k+ entries in our collected Chinese policy corpus. Experiments on lexical, embedding and fine-tuned bi-encoder models show the effectiveness of our proposed CSPRD yet also suggests ample potential for improvement. Our best performing baseline achieves 56.1% MRR@10, 28.5% NDCG@10, 37.5% Recall@10 and 80.6% Precision@10 on dev set.

Zhong Wang, Zengyu Wan, Han Han et al.

Event-based eye tracking has shown great promise with the high temporal resolution and low redundancy provided by the event camera. However, the diversity and abruptness of eye movement patterns, including blinking, fixating, saccades, and smooth pursuit, pose significant challenges for eye localization. To achieve a stable event-based eye-tracking system, this paper proposes a bidirectional long-term sequence modeling and time-varying state selection mechanism to fully utilize contextual temporal information in response to the variability of eye movements. Specifically, the MambaPupil network is proposed, which consists of the multi-layer convolutional encoder to extract features from the event representations, a bidirectional Gated Recurrent Unit (GRU), and a Linear Time-Varying State Space Module (LTV-SSM), to selectively capture contextual correlation from the forward and backward temporal relationship. Furthermore, the Bina-rep is utilized as a compact event representation, and the tailor-made data augmentation, called as Event-Cutout, is proposed to enhance the model's robustness by applying spatial random masking to the event image. The evaluation on the ThreeET-plus benchmark shows the superior performance of the MambaPupil, which secured the 1st place in CVPR'2024 AIS Event-based Eye Tracking challenge.

Ke Hong, Lufang Chen, Zhong Wang et al.

Existing large language model (LLM) serving systems fall into two categories: 1) a unified system where prefill phase and decode phase are co-located on the same GPU, sharing the unified computational resource and storage, and 2) a disaggregated system where the two phases are disaggregated to different GPUs. The design of the disaggregated system addresses the latency interference and sophisticated scheduling issues in the unified system but leads to storage challenges including 1) replicated weights for both phases that prevent flexible deployment, 2) KV cache transfer overhead between the two phases, 3) storage imbalance that causes substantial wasted space of the GPU capacity, and 4) suboptimal resource adjustment arising from the difficulties in migrating KV cache. Such storage inefficiency delivers poor serving performance under high request rates. In this paper, we identify that the advantage of the disaggregated system lies in the disaggregated computation, i.e., partitioning the computational resource to enable the asynchronous computation of two phases. Thus, we propose a novel LLM serving system, semi-PD, characterized by disaggregated computation and unified storage. In semi-PD, we introduce a computation resource controller to achieve disaggregated computation at the streaming multi-processor (SM) level, and a unified memory manager to manage the asynchronous memory access from both phases. semi-PD has a low-overhead resource adjustment mechanism between the two phases, and a service-level objective (SLO) aware dynamic partitioning algorithm to optimize the SLO attainment. Compared to state-of-the-art systems, semi-PD maintains lower latency at higher request rates, reducing the average end-to-end latency per request by 1.27-2.58x on DeepSeek series models, and serves 1.55-1.72x more requests adhering to latency constraints on Llama series models.