arXiv subjects

Juncheng Zhang

Publications and source records attributed to Juncheng Zhang.

10 recordsLinked to original sources

arXiv2026

LUMEN: Coordinated Failure Recovery for Distributed LLM Serving

Modern large language model (LLM) serving clusters distribute inference requests across multiple worker processes on different GPUs, but failures are prevalent at scale. When a worker fails, the cluster simultaneously loses the failed worker's GPU-resident key-value (KV) caches and serving capacity, leaving surviving workers to absorb the redirected traffic while re-running interrupted requests from scratch. Existing fault-tolerant systems either restart interrupted requests from scratch or restore KV caches from checkpoints stored on a fixed neighboring worker, but both approaches route recovery work without considering current cluster load and leave the recovering worker idle during model reload. We present LUMEN, a fault-tolerant LLM serving system that treats recovery as a load-aware coordination problem across three decision points: checkpoint placement before failures, interrupted-request distribution at failure time, and serving capacity restoration during model reload. We evaluate LUMEN using both prototype experiments and large-scale simulations and demonstrate significant improvements in serving and recovery times.

cs.DC↗

arXiv2026

Leveraging I/O Stalls for Efficient Scheduling in ANNS

Disk-based graph indexes for approximate nearest neighbor search (ANNS) must serve latency-sensitive queries and throughput-demanding updates concurrently. We observe that over 40% of search-thread CPU time is spent stalling on disk I/O; such idle cycles are invisible to thread-level scheduling yet available for other work. We present LIOS(Leverage I/O Stall), a framework that executes index updates inside search-side I/O stall windows. LIOS introduces three techniques: (i) splitting each update into resumable subtasks small enough to fit within a single stall window; (ii) bounding the expected overrun of update subtasks to a given threshold; and (iii) dynamically adjusting the fraction of idle time devoted to updates to drive end-to-end search latency degradation toward a user-specified target. We integrate LIOS into two update-optimized ANNS systems, FreshDiskANN and OdinANN. LIOS achieves speedups of up to 2.68$\times$ in insertion and 2.18$\times$ in deletion, with search latency degradation maintained near the user-specified target.

cs.DB↗

arXiv2026

AdaCluster: Adaptive Query-Key Clustering for Sparse Attention in Video Generation

Video diffusion transformers (DiTs) suffer from prohibitive inference latency due to quadratic attention complexity. Existing sparse attention methods either overlook semantic similarity or fail to adapt to heterogeneous token distributions across layers, leading to model performance degradation. We propose AdaCluster, a training-free adaptive clustering framework that accelerates the generation of DiTs while preserving accuracy. AdaCluster applies an angle-similarity-preserving clustering method to query vectors for higher compression, and designs a euclidean-similarity-preserving clustering method for keys, covering cluster number assignment, threshold-wise adaptive clustering, and efficient critical cluster selection. Experiments on CogVideoX-2B, HunyuanVideo, and Wan-2.1 on one A40 GPU demonstrate up to 1.67-4.31x speedup with negligible quality degradation.

cs.CV↗

arXiv2026

Decoupling Vector Data and Index Storage for Space Efficiency

Managing large-scale vector datasets with disk-resident graph approximate nearest neighbor search (ANNS) systems incurs substantial storage overhead due to the co-location of vector data and auxiliary index metadata, which prevents the storage layer from exploiting their distinct compressibility. We present COMPASS, a component-aware compressed storage framework for disk-resident graph vector search. Leveraging data-index decoupling as a foundation, COMPASS losslessly compresses each component according to its distinct compressibility characteristics, thereby significantly reducing storage space. It further adapts the search and update paths to preserve their performance under compressed storage layouts. Evaluation on real-world public and proprietary billion-scale datasets shows that COMPASS reduces storage space by up to 58.7%, while delivering improved or competitive search and update performance compared to state-of-the-art disk-resident graph ANNS systems.

cs.DB↗

arXiv2025

CentroidKV: Efficient Long-Context LLM Inference via KV Cache Clustering

Large language models (LLMs) with extended context windows have become increasingly prevalent for tackling complex tasks. However, the substantial Key-Value (KV) cache required for long-context LLMs poses significant deployment challenges. Existing approaches either discard potentially critical information needed for future generations or offer limited efficiency gains due to high computational overhead. In this paper, we introduce CentroidKV, a simple yet effective framework for online KV cache clustering. Our approach is based on the observation that key states exhibit high similarity along the sequence dimension. To enable efficient clustering, we divide the sequence into chunks and propose Chunked Soft Matching, which employs an alternating partition strategy within each chunk and identifies clusters based on similarity. CentroidKV then merges the KV cache within each cluster into a single centroid. Additionally, we provide a theoretical analysis of the computational complexity and the optimality of the intra-chunk partitioning strategy. Extensive experiments across various models and long-context benchmarks demonstrate that CentroidKV achieves up to 75% reduction in KV cache memory usage while maintaining comparable model performance. Moreover, with minimal computational overhead, CentroidKV accelerates the decoding stage of inference by up to $1.92\times$ and increases the serving throughput by up to $4\times$.

cs.CL↗

arXiv2025

HATA: Trainable and Hardware-Efficient Hash-Aware Top-k Attention for Scalable Large Model Inference

Large Language Models (LLMs) have emerged as a pivotal research area, yet the attention module remains a critical bottleneck in LLM inference, even with techniques like KVCache to mitigate redundant computations. While various top-$k$ attention mechanisms have been proposed to accelerate LLM inference by exploiting the inherent sparsity of attention, they often struggled to strike a balance between efficiency and accuracy. In this paper, we introduce HATA (Hash-Aware Top-$k$ Attention), a novel approach that systematically integrates low-overhead learning-to-hash techniques into the Top-$k$ attention process. Different from the existing top-k attention methods which are devoted to seeking an absolute estimation of qk score, typically with a great cost, HATA maps queries and keys into binary hash codes, and acquires the relative qk score order with a quite low cost, which is sufficient for realizing top-k attention. Extensive experiments demonstrate that HATA achieves up to 7.2$\times$ speedup compared to vanilla full attention while maintaining model accuracy. In addition, HATA outperforms the state-of-the-art top-$k$ attention methods in both accuracy and efficiency across multiple mainstream LLM models and diverse tasks. HATA is open source at https://github.com/gpzlx1/HATA.

cs.LG↗

arXiv2024

Design and Process Analysis of a Split-Gate Trench Power MOSFET with Bottom-Trench Hk-Pillar Superjunction for Enhanced Performance

In this paper, we propose a simulation-based novel Split-Gate Trench MOSFET structure with an optimized fabrication process to enhance power efficiency, switching speed, and thermal stability for high-performance semiconductor applications. Integrating high-k pillars with superjunction structures beneath the split gate enhancing breakdown performance by reducing critical field intensity by up to 35%, the device achieves a 15% improvement in Figures of Merit (FOMs) for BV2/Ron,sp. Dynamic testing reveals approximately a 25% reduction in both input and output capacitance, as well as gate-to-drain charge (QGD). This reduction, coupled with an approximately 40% improvement in Baliga's High-Frequency Figure of Merit (BHFFOM) and over 20% increase in the New High-Frequency Figure of Merit (NHFFOM), underscores the design's suitability for high-speed, high-efficiency power electronics. Simulations examining the effects of high-k pillar depth indicate that an optimal depth of 3.5 um achieves a balanced performance between BV and Ron,sp. The influence of high-k materials on BT-Hk-SJ MOSFET performance was investigated by comparing hafnium dioxide (HfO2), nitride, and oxynitride. Among these, HfO2 demonstrated optimal performance across static, dynamic, and diode characteristics due to its high dielectric constant, while material choice had minimal impact, with variations kept within 5%.

physics.app-ph↗

arXiv2020

Real-MFF: A Large Realistic Multi-focus Image Dataset with Ground Truth

Multi-focus image fusion, a technique to generate an all-in-focus image from two or more partially-focused source images, can benefit many computer vision tasks. However, currently there is no large and realistic dataset to perform convincing evaluation and comparison of algorithms in multi-focus image fusion. Moreover, it is difficult to train a deep neural network for multi-focus image fusion without a suitable dataset. In this letter, we introduce a large and realistic multi-focus dataset called Real-MFF, which contains 710 pairs of source images with corresponding ground truth images. The dataset is generated by light field images, and both the source images and the ground truth images are realistic. To serve as both a well-established benchmark for existing multi-focus image fusion algorithms and an appropriate training dataset for future development of deep-learning-based methods, the dataset contains a variety of scenes, including buildings, plants, humans, shopping malls, squares and so on. We also evaluate 10 typical multi-focus algorithms on this dataset for the purpose of illustration.

cs.CV↗

arXiv2019

An α-Matte Boundary Defocus Model Based Cascaded Network for Multi-focus Image Fusion

Capturing an all-in-focus image with a single camera is difficult since the depth of field of the camera is usually limited. An alternative method to obtain the all-in-focus image is to fuse several images focusing at different depths. However, existing multi-focus image fusion methods cannot obtain clear results for areas near the focused/defocused boundary (FDB). In this paper, a novel α-matte boundary defocus model is proposed to generate realistic training data with the defocus spread effect precisely modeled, especially for areas near the FDB. Based on this α-matte defocus model and the generated data, a cascaded boundary aware convolutional network termed MMF-Net is proposed and trained, aiming to achieve clearer fusion results around the FDB. More specifically, the MMF-Net consists of two cascaded sub-nets for initial fusion and boundary fusion, respectively; these two sub-nets are designed to first obtain a guidance map of FDB and then refine the fusion near the FDB. Experiments demonstrate that with the help of the new α-matte boundary defocus model, the proposed MMF-Net outperforms the state-of-the-art methods both qualitatively and quantitatively.

cs.CV↗

arXiv2019

Boundary Aware Multi-Focus Image Fusion Using Deep Neural Network

Since it is usually difficult to capture an all-in-focus image of a 3D scene directly, various multi-focus image fusion methods are employed to generate it from several images focusing at different depths. However, the performance of existing methods is barely satisfactory and often degrades for areas near the focused/defocused boundary (FDB). In this paper, a boundary aware method using deep neural network is proposed to overcome this problem. (1) Aiming to acquire improved fusion images, a 2-channel deep network is proposed to better extract the relative defocus information of the two source images. (2) After analyzing the different situations for patches far away from and near the FDB, we use two networks to handle them respectively. (3) To simulate the reality more precisely, a new approach of dataset generation is designed. Experiments demonstrate that the proposed method outperforms the state-of-the-art methods, both qualitatively and quantitatively.

cs.CV↗