arXiv subjects

Yutong Lu

Publications and source records attributed to Yutong Lu.

At least 37 records · Page 2Linked to original sources

arXiv2026

AscendKernelGen: A Systematic Study of LLM-Based Kernel Generation for Neural Processing Units

To meet the ever-increasing demand for computational efficiency, Neural Processing Units (NPUs) have become critical in modern AI infrastructure. However, unlocking their full potential requires developing high-performance compute kernels using vendor-specific Domain-Specific Languages (DSLs), a task that demands deep hardware expertise and is labor-intensive. While Large Language Models (LLMs) have shown promise in general code generation, they struggle with the strict constraints and scarcity of training data in the NPU domain. Our preliminary study reveals that state-of-the-art general-purpose LLMs fail to generate functional complex kernels for Ascend NPUs, yielding a near-zero success rate. To address these challenges, we propose AscendKernelGen, a generation-evaluation integrated framework for NPU kernel development. We introduce Ascend-CoT, a high-quality dataset incorporating chain-of-thought reasoning derived from real-world kernel implementations, and KernelGen-LM, a domain-adaptive model trained via supervised fine-tuning and reinforcement learning with execution feedback. Furthermore, we design NPUKernelBench, a comprehensive benchmark for assessing compilation, correctness, and performance across varying complexity levels. Experimental results demonstrate that our approach significantly bridges the gap between general LLMs and hardware-specific coding. Specifically, the compilation success rate on complex Level-2 kernels improves from 0% to 95.5% (Pass@10), while functional correctness achieves 64.3% compared to the baseline's complete failure. These results highlight the critical role of domain-specific reasoning and rigorous evaluation in automating accelerator-aware code generation. AscendKernGen is available at https://huggingface.co/AscendKernelGen and https://github.com/weich97/NPUKernelBench.

cs.AI↗

arXiv2026

Breaking the Training Barrier of Billion-Parameter Universal Machine Learning Interatomic Potentials

Universal Machine Learning Interatomic Potentials (uMLIPs), pre-trained on massively diverse datasets encompassing inorganic materials and organic molecules across the entire periodic table, serve as foundational models for quantum-accurate physical simulations. However, uMLIP training requires second-order derivatives, which lack corresponding parallel training frameworks; moreover, scaling to the billion-parameter regime causes explosive growth in computation and communication overhead, making its training a tremendous challenge. We introduce MatRIS-MoE, a billion-parameter Mixture-of-Experts model built upon invariant architecture, and {Janus}, a pioneering high-dimensional distributed training framework for uMLIPs with hardware-aware optimizations. Deployed across two Exascale supercomputers, our code attains a peak performance of 1.2/1.0 EFLOPS (24\%/{35.5\%} of theoretical peak) in single precision at over 90\% parallel efficiency, compressing the training of billion-parameter uMLIPs from weeks to hours. This work establishes a new high-water mark for AI-for-Science (AI4S) foundation models at Exascale and provides essential infrastructure for rapid scientific discovery.

cs.DC↗

arXiv2026

AIM: Asymmetric Information Masking for Visual Question Answering Continual Learning

In continual visual question answering (VQA), existing Continual Learning (CL) methods are mostly built for symmetric, unimodal architectures. However, modern Vision-Language Models (VLMs) violate this assumption, as their trainable components are inherently asymmetric. This structural mismatch renders VLMs highly prone to catastrophic forgetting when learning from continuous data streams. Specifically, the asymmetry causes standard global regularization to favor the massive language decoder during optimization, leaving the smaller but critical visual projection layers highly vulnerable to interference. Consequently, this localized degradation leads to a severe loss of compositional reasoning capabilities. To address this, we propose Asymmetric Information Masking (AIM), which balances stability and plasticity by applying targeted masks based on modality-specific sensitivity. Experiments on VQA v2 and GQA under continual VQA settings show that AIM achieves state-of-the-art performance in both Average Performance (AP) and Average Forgetting (AF), while better preserving generalization to novel skill-concept compositions.

cs.CV↗

arXiv2026

GTPBD-MM: A Global Terraced Parcel and Boundary Dataset with Multi-Modality

Agricultural parcel extraction plays an important role in remote sensing-based agricultural monitoring, supporting parcel surveying, precision management, and ecological assessment. However, existing public benchmarks mainly focus on regular and relatively flat farmland scenes. In contrast, terraced parcels in mountainous regions exhibit stepped terrain, pronounced elevation variation, irregular boundaries, and strong cross-regional heterogeneity, making parcel extraction a more challenging problem that jointly requires visual recognition, semantic discrimination, and terrain-aware geometric understanding. Although recent studies have advanced visual parcel benchmarks and image-text farmland understanding, a unified benchmark for complex terraced parcel extraction under aligned image-text-DEM settings remains absent. To fill this gap, we present GTPBD-MM, the first multimodal benchmark for global terraced parcel extraction. Built upon GTPBD, GTPBD-MM integrates high-resolution optical imagery, structured text descriptions, and DEM data, and supports systematic evaluation under Image-only, Image+Text, and Image+Text+DEM settings. We further propose Elevation and Text guided Terraced parcel network (ETTerra), a multimodal baseline for terraced parcel delineation. Extensive experiments demonstrate that textual semantics and terrain geometry provide complementary cues beyond visual appearance alone, yielding more accurate, coherent, and structurally consistent delineation results in complex terraced scenes.

cs.CV↗

arXiv2026

Beyond Few-Step Inference: Accelerating Video Diffusion Transformer Model Serving with Inter-Request Caching Reuse

Video Diffusion Transformer (DiT) models are a dominant approach for high-quality video generation but suffer from high inference cost due to iterative denoising. Existing caching approaches primarily exploit similarity within the diffusion process of a single request to skip redundant denoising steps. In this paper, we introduce Chorus, a caching approach that leverages similarity across requests to accelerate video diffusion model serving. Chorus achieves up to 45\% speedup on industrial 4-step distilled models, where prior intra-request caching approaches are ineffective. Particularly, Chorus employs a three-stage caching strategy along the denoising process. Stage 1 performs full reuse of latent features from similar requests. Stage 2 exploits inter-request caching in specific latent regions during intermediate denoising steps. This stage is combined with Token-Guided Attention Amplification to improve semantic alignment between the generated video and the conditional prompts, thereby extending the applicability of full reuse to later denoising steps.

cs.CV↗

arXiv2026

AGCD: Agent-Guided Cross-Modal Decoding for Weather Forecasting

Accurate weather forecasting is more than grid-wise regression: it must preserve coherent synoptic structures and physical consistency of meteorological fields, especially under autoregressive rollouts where small one-step errors can amplify into structural bias. Existing physics-priors approaches typically impose global, once-for-all constraints via architectures, regularization, or NWP coupling, offering limited state-adaptive and sample-specific controllability at deployment. To bridge this gap, we propose Agent-Guided Cross-modal Decoding (AGCD), a plug-and-play decoding-time prior-injection paradigm that derives state-conditioned physics-priors from the current multivariate atmosphere and injects them into forecasters in a controllable and reusable way. Specifically, We design a multi-agent meteorological narration pipeline to generate state-conditioned physics-priors, utilizing MLLMs to extract various meteorological elements effectively. To effectively apply the priors, AGCD further introduce cross-modal region interaction decoding that performs region-aware multi-scale tokenization and efficient physics-priors injection to refine visual features without changing the backbone interface. Experiments on WeatherBench demonstrate consistent gains for 6-hour forecasting across two resolutions (5.625 degree and 1.40625 degree) and diverse backbones (generic and weather-specialized), including strictly causal 48-hour autoregressive rollouts that reduce early-stage error accumulation and improve long-horizon stability.

cs.AI↗

arXiv2026

Clinical-Injection Transformer with Domain-Adapted MAE for Lupus Nephritis Prognosis Prediction

Lupus nephritis (LN) is a severe complication of systemic lupus erythematosus that affects pediatric patients with significantly greater severity and worse renal outcomes compared to adults. Despite the urgent clinical need, predicting pediatric LN prognosis remains unexplored in computational pathology. Furthermore, the only existing histopathology-based approach for LN relies on multiple costly staining protocols and fails to integrate complementary clinical data. To address these gaps, we propose the first multimodal computational pathology framework for three-class treatment response prediction (complete remission, partial response, and no response) in pediatric LN, utilizing only routine PAS-stained biopsies and structured clinical data. Our framework introduces two key methodological innovations. First, a Clinical-Injection Transformer (CIT) embeds clinical features as condition tokens into patch-level self-attention, facilitating implicit and bidirectional cross-modal interactions within a unified attention space. Second, we design a decoupled representation-knowledge adaptation strategy using a domain-adapted Masked Autoencoder (MAE). This strategy explicitly separates self-supervised morphological feature learning from pathological knowledge extraction. Additionally, we introduce a multi-granularity morphological type injection mechanism to bridge distilled classification knowledge with downstream prognostic predictions at both the instance and patient levels. Evaluated on a cohort of 71 pediatric LN patients with KDIGO-standardized labels, our method achieves a three-class accuracy of 90.1% and an AUC of 89.4%, demonstrating its potential as a highly accurate and cost-effective prognostic tool.

eess.IV↗

arXiv2026

VeriSciQA: An Auto-Verified Dataset for Scientific Visual Question Answering

Large Vision-Language Models (LVLMs) show promise for scientific applications, yet open-source models still struggle with Scientific Visual Question Answering (SVQA), namely answering questions about figures from scientific papers. A key bottleneck is the lack of public, large-scale, high-quality SVQA datasets. Although recent work uses LVLMs to synthesize data at scale, we identify systematic errors in their resulting QA pairs, stemming from LVLMs' inherent limitations and information asymmetry between figures and text. To address these challenges, we propose a Cross-Modal verification framework that generates questions and answers purely from figure-citing paragraphs, then verifies them against the figures themselves, leveraging the inherent text-figure alignment in scientific papers to filter out erroneous QA pairs. We instantiate this framework to curate VeriSciQA, a dataset of 20,272 QA pairs spanning 20 scientific domains and 12 figure types. Difficulty assessment reveals a notable accuracy gap between the best open-source model (65%) and the best proprietary model (80.5%), demonstrating room for improvement. Moreover, models fine-tuned on VeriSciQA achieve consistent improvements on SVQA benchmarks, with performance gains that scale with data size, surpassing models trained on existing datasets. Human evaluation further validates the improved quality of VeriSciQA. These results demonstrate that continued data expansion via our scalable framework can further advance SVQA capability in the open-source community. Our dataset is publicly available at https://huggingface.co/datasets/datajuicer/VeriSciQA.

cs.CV↗

arXiv2026

Matrix-PIC: Harnessing Matrix Outer-product for High-Performance Particle-in-Cell Simulations

Particle-in-Cell (PIC) simulations spend most of their execution time on particle--grid interactions, where fine-grained atomic updates become a major bottleneck on traditional many-core CPUs. Recent CPU architectures integrate specialized Matrix Processing Units (MPUs) that efficiently support matrix outer-product operations, offering new opportunities to overcome this limitation. Leveraging this architectural shift, this work focuses on redesigning the current deposition step of PIC simulations under a matrix-centric execution model. We present MatrixPIC, the first holistic co-design of the deposition kernel, data layout, and incremental particle sorting tailored to the hybrid MPU--VPU SIMD model on modern CPUs. MatrixPIC introduces: (i)~a block-matrix formulation of the current deposition algorithm that maps naturally to MPU outer-product primitives; (ii)~a hybrid execution pipeline that combines MPU-based high-density accumulation with VPU-based data preparation and control flow; and (iii)~an $O(1)$-amortized incremental sorter based on a gapped packed-memory array to preserve data locality for efficient MPU execution. Evaluated on a next-generation HPC platform, MatrixPIC achieves significant performance gains. In Laser-Wakefield Acceleration (LWFA) simulations, it delivers up to $2.63\times$ speedup in total runtime. For third-order deposition, the core kernel is accelerated by $8.7\times$ over the baseline and $2.0\times$ over the best hand-optimized VPU implementation. Moreover, MatrixPIC reaches $83.08\%$ of theoretical CPU peak performance, nearly $2.8\times$ higher than a highly optimized CUDA kernel on a data center GPU. These results demonstrate the effectiveness of matrix-oriented co-design for accelerating PIC simulations on emerging CPU architectures.

cs.DC↗

arXiv2025

Equivariant Diffusion for Crystal Structure Prediction

In addressing the challenge of Crystal Structure Prediction (CSP), symmetry-aware deep learning models, particularly diffusion models, have been extensively studied, which treat CSP as a conditional generation task. However, ensuring permutation, rotation, and periodic translation equivariance during diffusion process remains incompletely addressed. In this work, we propose EquiCSP, a novel equivariant diffusion-based generative model. We not only address the overlooked issue of lattice permutation equivariance in existing models, but also develop a unique noising algorithm that rigorously maintains periodic translation equivariance throughout both training and inference processes. Our experiments indicate that EquiCSP significantly surpasses existing models in terms of generating accurate structures and demonstrates faster convergence during the training process.

cond-mat.mtrl-sci↗

arXiv2025

Partial Cross-Compilation and Mixed Execution for Accelerating Dynamic Binary Translation

With the growing diversity of instruction set architectures (ISAs), cross-ISA program execution has become common. Dynamic binary translation (DBT) is the main solution but suffers from poor performance. Cross-compilation avoids emulation costs but is constrained by an "all-or-nothing" model-programs are either fully cross-compiled or entirely emulated. Complete cross-compilation is often unfeasible due to ISA-specific code or missing dependencies, leaving programs with high emulation overhead. We propose a hybrid execution system that combines compilation and emulation, featuring a selective function offloading mechanism. This mechanism establishes cross-environment calling channels, offloading eligible functions to the host for native execution to reduce DBT overhead. Key optimizations address offloading costs, enabling efficient hybrid operation. Built on LLVM and QEMU, the system works automatically for both applications and libraries. Evaluations show it achieves up to 13x speedups over existing DBT, with strong practical value.

cs.AR↗

arXiv2025

PolyKAN: Efficient Fused GPU Operators for Polynomial Kolmogorov-Arnold Network Variants

Kolmogorov-Arnold Networks (KANs) promise higher expressive capability and stronger interpretability than Multi-Layer Perceptron, particularly in the domain of AI for Science. However, practical adoption has been hindered by low GPU utilization of existing parallel implementations. To address this challenge, we present a GPU-accelerated operator library, named PolyKAN which is the first general open-source implementation of KAN and its variants. PolyKAN fuses the forward and backward passes of polynomial KAN layers into a concise set of optimized CUDA kernels. Four orthogonal techniques underpin the design: (i) \emph{lookup-table} with linear interpolation that replaces runtime expensive math-library functions; (ii) \emph{2D tiling} to expose thread-level parallelism with preserving memory locality; (iii) a \emph{two-stage reduction} scheme converting scattered atomic updates into a single controllable merge step; and (iv) \emph{coefficient-layout reordering} yielding unit-stride reads under the tiled schedule. Using a KAN variant, Chebyshev KAN, as a case-study, PolyKAN delivers $1.2$--$10\times$ faster inference and $1.4$--$12\times$ faster training than a Triton + cuBLAS baseline, with identical accuracy on speech, audio-enhancement, and tabular-regression workloads on both highend GPU and consumer-grade GPU.

cs.DC↗

arXiv2025

Boosting LLM Serving through Spatial-Temporal GPU Resource Sharing

Modern LLM serving systems confront inefficient GPU utilization due to the fundamental mismatch between compute-intensive prefill and memory-bound decode phases. While current practices attempt to address this by organizing these phases into hybrid batches, such solutions create an inefficient tradeoff that sacrifices either throughput or latency, leaving substantial GPU resources underutilized. We identify two key root causes: 1) the prefill phase suffers from suboptimal compute utilization due to wave quantization and attention bottlenecks. 2) hybrid batches disproportionately prioritize latency over throughput, resulting in wasted compute and memory bandwidth. To mitigate the issues, we present Bullet, a novel spatial-temporal orchestration system that eliminates these inefficiencies through precise phase coordination. Bullet enables concurrent execution of prefill and decode phases, while dynamically provisioning GPU resources using real-time performance modeling. By integrating SLO-aware scheduling and adaptive resource allocation, Bullet maximizes utilization without compromising latency targets. Experimental evaluations on real-world workloads demonstrate that Bullet delivers 1.26x average throughput gains (up to 1.55x) over state-of-the-arts, while consistently meeting latency constraints.

cs.DC↗

arXiv2025

HPC-AI Coupling Methodology for Scientific Applications

Artificial intelligence (AI) technologies have fundamentally transformed numerical-based high-performance computing (HPC) applications with data-driven approaches and endeavored to address existing challenges, e.g. high computational intensity, in various scientific domains. In this study, we explore the scenarios of coupling HPC and AI (HPC-AI) in the context of emerging scientific applications, presenting a novel methodology that incorporates three patterns of coupling: surrogate, directive, and coordinate. Each pattern exemplifies a distinct coupling strategy, AI-driven prerequisite, and typical HPC-AI ensembles. Through case studies in materials science, we demonstrate the application and effectiveness of these patterns. The study highlights technical challenges, performance improvements, and implementation details, providing insight into promising perspectives of HPC-AI coupling. The proposed coupling patterns are applicable not only to materials science but also to other scientific domains, offering valuable guidance for future HPC-AI ensembles in scientific discovery.

cs.CE↗

arXiv2025

TD-Pipe: Temporally-Disaggregated Pipeline Parallelism Architecture for High-Throughput LLM Inference

As the model size continuously increases, pipeline parallelism shows great promise in throughput-oriented LLM inference due to its low demand on communications. However, imbalanced pipeline workloads and complex data dependencies in the prefill and decode phases result in massive pipeline bubbles and further severe performance reduction. To better exploit the pipeline parallelism for high-throughput LLM inference, we propose TD-Pipe, with the key idea lies in the temporally-disaggregated pipeline parallelism architecture. Specifically, this architecture disaggregates the prefill and decode phases in the temporal dimension, so as to eliminate pipeline bubbles caused by the phase switching. TD-Pipe identifies potential issues of exploiting the novel architecture and provides solutions. First, a hierarchy-controller structure is used to better coordinate devices in pipeline parallelism by decoupling the scheduling from execution. Second, the AI-based greedy prefill approach aggressively performs more prefills by predicting the output length and simulating the memory usage. Third, the inter-batch work stealing approach dynamically balances decode phase workloads between different batches to reduce bubbles. Forth, the spatial-temporal intensity comparison approach determines the optimal switch from decode to prefill by comparing the performance drop from reduced computational intensity with that from phase switching bubbles. Extensive experiments show that TD-Pipe effectively increases the throughput of LLM inference by up to 1.91x over the existing tensor parallel approach and 2.73x over the existing pipeline parallel approach on GPU nodes with only PCIe interconnection.

cs.DC↗

arXiv2025

Ghidorah: Fast LLM Inference on Edge with Speculative Decoding and Hetero-Core Parallelism

In-situ LLM inference on end-user devices has gained significant interest due to its privacy benefits and reduced dependency on external infrastructure. However, as the decoding process is memory-bandwidth-bound, the diverse processing units in modern end-user devices cannot be fully exploited, resulting in slow LLM inference. This paper presents Ghidorah, a LLM inference system for end-user devices with the unified memory architecture. The key idea of Ghidorah can be summarized in two steps: 1) leveraging speculative decoding approaches to enhance parallelism, and 2) ingeniously distributing workloads across multiple heterogeneous processing units to maximize computing power utilization. Ghidorah includes the hetero-core model parallelism (HCMP) architecture and the architecture-aware profiling (ARCA) approach. The HCMP architecture guides partitioning by leveraging the unified memory design of end-user devices and adapting to the hybrid computational demands of speculative decoding. The ARCA approach is used to determine the optimal speculative strategy and partitioning strategy, balancing acceptance rate with parallel capability to maximize the speedup. Additionally, we optimize sparse computation on ARM CPUs. Experimental results show that Ghidorah can achieve up to 7.6x speedup in the dominant LLM decoding phase compared to the sequential decoding approach in NVIDIA Jetson NX.

cs.DC↗

arXiv2025

gLLM: Global Balanced Pipeline Parallelism System for Distributed LLM Serving with Token Throttling

Pipeline parallelism has emerged as a predominant approach for deploying large language models (LLMs) across distributed nodes, owing to its lower communication overhead compared to tensor parallelism. While demonstrating high throughput in request serving, pipeline parallelism often suffers from performance limitations caused by pipeline bubbles, which are primarily resulted from imbalanced computation delays across batches. Existing methods like Sarathi-Serve attempt to address this through hybrid scheduling of chunked prefill and decode tokens using a fixed token budget. However, such methods may experience significant fluctuations due to either insufficient prefill tokens or uneven distribution of decode tokens, ultimately leading to computational imbalance. To overcome these inefficiencies, we present gLLM, a globally balanced pipeline parallelism system incorporating Token Throttling to effectively mitigate the pipeline bubbles. Our Token Throttling mechanism is a fine-grained scheduling policy that independently regulates the quantities of prefill and decode tokens, thus enabling balanced computation by leveraging global information from the inference system. Specifically, for decode tokens, gLLM maintains near-consistent token count across processing batches. For prefill tokens, it dynamically adjusts batch sizes based on both total pending tokens and the memory utilization rates of key-value cache (KV cache). Furthermore, gLLM runtime adopts an asynchronous execution and message passing architecture specifically optimized for pipeline parallelism characteristics. Experimental evaluations with representative LLMs show that gLLM achieves significant performance improvements, delivering 11% to 398% higher maximum throughput compared to state-of-the-art pipeline or tensor parallelism systems, while simultaneously maintaining lower latency.

cs.DC↗

arXiv2025

EcoServe: Enabling Cost-effective LLM Serving with Proactive Intra- and Inter-Instance Orchestration

Existing LLM serving strategies can be categorized based on whether prefill and decode phases are disaggregated: non-disaggregated (NoDG) or fully disaggregated (FuDG). However, the NoDG strategy leads to strong prefill-decode interference and the FuDG strategy highly relies on high-performance interconnects, making them less cost-effective. We introduce EcoServe, a system that enables cost-effective LLM serving on clusters with commodity interconnects. EcoServe is built on the partially disaggregated (PaDG) strategy, applying temporal disaggregation and rolling activation for proactive intra- and inter-instance scheduling. It first disaggregates the prefill and decode phases along the time dimension within a single instance to mitigate inter-phase interference and enhance throughput. Next, it coordinates multiple instances and cyclically activates them to ensure the continuous availability of prefill processing, thereby improving latency. Thus, EcoServe's basic serving unit is the macro instance, within which multiple instances collaborate. It further integrates an adaptive scheduling algorithm to route requests in a macro instance and a mitosis scaling approach to enable fine-grained capacity scaling. Beyond delivering high goodput, EcoServe excels in load balancing, hardware cost, parallelism compatibility, and even engineering simplicity compared to existing solutions. When serving 30B- and 70B-scale models on a production-level cluster with 32 NVIDIA L20 GPUs using commodity Ethernet, EcoServe averagely improves goodput by 82.49%, 86.17%, 122.76%, and 126.96% over four representative NoDG and FuDG systems.

cs.DC↗