arXiv subjects

Feiyi Wang

Publications and source records attributed to Feiyi Wang.

At least 19 recordsLinked to original sources

arXiv2026

Piper: Efficient Large-Scale MoE Training via Resource Modeling and Pipelined Hybrid Parallelism

Frontier models increasingly adopt Mixture-of-Experts (MoE) architectures to achieve large-model performance at reduced cost. However, training MoE models on HPC platforms is hindered by large memory footprints, frequent large-scale communication across heterogeneous networks, and severe workload imbalance. To characterize these challenges, we develop a mathematical model that quantifies memory, compute, and communication requirements for MoE configurations under various parallelization schemes, verified through micro-benchmarking, code instrumentation, and hardware profiling. Our analysis identifies performance bottlenecks: all-to-all latency at scale from expert parallelism, insufficient compute-communication overlap, low GPU utilization from imbalanced skinny GEMMs, and the absence of platform-aware hybrid parallelization strategies. To address these, we introduce Piper, a framework that leverages resource modeling to identify efficient training strategies for MoE models on target HPC platforms, applying pipeline parallelism with optimized schedules. Piper achieves 2-3.5X higher MFU than state-of-the-art frameworks such as X-MoE, and a novel all-to-all algorithm delivers 1.2-9X bandwidth over vendor implementation.

cs.DC↗

arXiv2026

Beyond Pixels: Introspective and Interactive Grounding for Visualization Agents

Vision-Language Models (VLMs) frequently misread values, hallucinate details, and confuse overlapping elements in charts. Current approaches rely solely on pixel interpretation, creating a Pixel-Only Bottleneck: agents treat interactive charts as static images, losing access to the structured specification that encodes exact values. We introduce Introspective and Interactive Visual Grounding (IVG), a framework that combines (1) spec-grounded introspection, which queries the underlying specification for deterministic evidence, with (2) view-grounded interaction, which manipulates the view to resolve visual ambiguity. To enable evaluation without VLM bias, we present iPlotBench, a benchmark of 500 interactive Plotly figures with 6,706 binary questions and ground-truth specifications. Experiments show that introspection improves data reconstruction fidelity, while the combination with interaction achieves the highest QA accuracy (0.81), with +6.7 % gains on overlapping geometries. We further demonstrate IVG in deployed agents that explore data autonomously and collaborate with human users in real time.

cs.CL↗

arXiv2026

FedSpy-LLM: Towards Scalable and Generalizable Data Reconstruction Attacks from Gradients on LLMs

Given the growing reliance on private data in training Large Language Models (LLMs), Federated Learning (FL) combined with Parameter-Efficient Fine-Tuning (PEFT) has garnered significant attention for enhancing privacy and efficiency. Despite FL's privacy benefits, prior studies have shown that private data can still be extracted from shared gradients. However, these studies, mainly on full-parameter model training, are limited to reconstructing small batches, short input sequences, and specific model architectures, such as encoder-based or decoder-based models. The reconstruction quality becomes even worse when dealing with gradients from PEFT methods. To fully understand the practical attack surface of federated LLMs, this paper proposes FedSpy-LLM, a scalable and generalizable data reconstruction attack designed to reconstruct training data with larger batch sizes and longer sequences while generalizing across diverse model architectures, even when PEFT methods are deployed for training. At the core of FedSpy-LLM is a novel gradient decomposition strategy that exploits the rank deficiency and subspace structure of gradients, enabling efficient token extraction while preserving key signal components at scale. This approach further mitigates the reconstruction challenges introduced by PEFT's substantial null space, ensuring robustness across encoder-based, decoder-based, and encoder-decoder model architectures. Additionally, by iteratively aligning each token's partial-sequence gradient with the full-sequence gradient, FedSpy-LLM ensures accurate token ordering in reconstructed sequences.

cs.CR↗

arXiv2026

Asymmetric Advantage Modulation Calibrates Entropy Dynamics in RLVR

Reinforcement learning with verifiable rewards (RLVR) has substantially improved the reasoning ability of large language models (LLMs), but it often suffers from \textit{restricted exploration}, where the policy rapidly concentrates on a narrow set of solutions. A common remedy is entropy regularization, which attempts to preserve exploration by increasing policy entropy. However, for LLM-RL, this intervention is highly sensitive to its coefficient, can introduce semantically weak uncertainty, and often yields limited accuracy gains. This motivates a more precise question: which entropy helps reasoning, and which entropy should be reduced? To study this, we parameterize the advantage estimator in Group Relative Policy Optimization (GRPO) into positive and negative outcome-conditioned channels and analyze their entropy dynamics. Our results show that positive-channel modulation raises \textit{productive entropy} associated with successful reasoning trajectories, while negative-channel modulation removes \textit{noisy entropy} associated with failed rollouts and reduces interference with correct paths. Guided by this channel-wise view, we propose \textbf{AsymGRPO}, which decouples the modulation strengths of positive and negative advantages. This enables flexible control over how the model updates across prompt difficulty levels, allowing stronger reinforcement of rare successes on harder prompts or stronger suppression of residual failures on easier prompts without forcing the two channels to share the same modulation strength. Experiments on five mathematical reasoning benchmarks show that AsymGRPO outperforms strong RLVR baselines, with consistent gains across model backbones.

cs.CL↗

arXiv2026

Tula: Optimizing Time, Cost, and Generalization in Distributed Large-Batch Training

Distributed training increases the number of batches processed per iteration either by scaling-out (adding more nodes) or scaling-up (increasing the batch-size). However, the largest configuration does not necessarily yield the best performance. Horizontal scaling introduces additional communication overhead, while vertical scaling is constrained by computation cost and device memory limits. Thus, simply increasing the batch-size leads to diminishing returns: training time and cost decrease initially but eventually plateaus, creating a knee-point in the time/cost versus batch-size pareto curve. The optimal batch-size therefore depends on the underlying model, data and available compute resources. Large batches also suffer from worse model quality due to the well-known generalization gap. In this paper, we present Tula, an online service that automatically optimizes time, cost, and convergence quality for large-batch training of convolutional models. It combines parallel-systems modeling with statistical performance prediction to identify the optimal batch-size. Tula predicts training time and cost within 7.5-14% error across multiple models, and achieves up to 20x overall speedup and improves test accuracy by 9% on average over standard large-batch training on various vision tasks, thus successfully mitigating the generalization gap and accelerating training at the same time.

cs.LG↗

arXiv2026

FLYING SERVING: On-the-Fly Parallelism Switching for Large Language Model Serving

Production LLM serving must simultaneously deliver high throughput, low latency, and sufficient context capacity under non-stationary traffic and mixed request requirements. Data parallelism (DP) maximizes throughput by running independent replicas, while tensor parallelism (TP) reduces per-request latency and pools memory for long-context inference. However, existing serving stacks typically commit to a static parallelism configuration at deployment; adapting to bursts, priorities, or long-context requests is often disruptive and slow. We present Flying Serving, a vLLM-based system that enables online DP-TP switching without restarting engine workers. Flying Serving makes reconfiguration practical by virtualizing the state that would otherwise force data movement: (i) a zero-copy Model Weights Manager that exposes TP shard views on demand, (ii) a KV Cache Adaptor that preserves request KV state across DP/TP layouts, (iii) an eagerly initialized Communicator Pool to amortize collective setup, and (iv) a deadlock-free scheduler that coordinates safe transitions under execution skew. Across three popular LLMs and realistic serving scenarios, Flying Serving improves performance by up to $4.79\times$ under high load and $3.47\times$ under low load while supporting latency- and memory-driven requests.

cs.DC↗

arXiv2026

Accelerating Large-Scale Dataset Distillation via Exploration-Exploitation Optimization

Dataset distillation compresses the original data into compact synthetic datasets, reducing training time and storage while retaining model performance, enabling deployment under limited resources. Although recent decoupling-based distillation methods enable dataset distillation at large scale, they continue to face an efficiency gap: optimization-based decoupling methods achieve higher accuracy but demand intensive computation, whereas optimization-free decoupling methods are efficient but sacrifice accuracy. To overcome this trade-off, we propose Exploration--Exploitation Distillation (E$^2$D), a simple, practical method that minimizes redundant computation through an efficient pipeline that begins with full-image initialization to preserve semantic integrity and feature diversity. It then uses a two-phase optimization strategy: an exploration phase that performs uniform updates and identifies high-loss regions, and an exploitation phase that focuses updates on these regions to accelerate convergence. We evaluate E$^2$D on large-scale benchmarks, surpassing the state-of-the-art on ImageNet-1K while being $18\times$ faster, and on ImageNet-21K, our method substantially improves accuracy while remaining $4.3\times$ faster. These results demonstrate that targeted, redundancy-reducing updates, rather than brute-force optimization, bridge the gap between accuracy and efficiency in large-scale dataset distillation. Code is available at https://github.com/ncsu-dk-lab/E2D.

cs.CV↗

arXiv2025

SciTrust 2.0: A Comprehensive Framework for Evaluating Trustworthiness of Large Language Models in Scientific Applications

Large language models (LLMs) have demonstrated transformative potential in scientific research, yet their deployment in high-stakes contexts raises significant trustworthiness concerns. Here, we introduce SciTrust 2.0, a comprehensive framework for evaluating LLM trustworthiness in scientific applications across four dimensions: truthfulness, adversarial robustness, scientific safety, and scientific ethics. Our framework incorporates novel, open-ended truthfulness benchmarks developed through a verified reflection-tuning pipeline and expert validation, alongside a novel ethics benchmark for scientific research contexts covering eight subcategories including dual-use research and bias. We evaluated seven prominent LLMs, including four science-specialized models and three general-purpose industry models, using multiple evaluation metrics including accuracy, semantic similarity measures, and LLM-based scoring. General-purpose industry models overall outperformed science-specialized models across each trustworthiness dimension, with GPT-o4-mini demonstrating superior performance in truthfulness assessments and adversarial robustness. Science-specialized models showed significant deficiencies in logical and ethical reasoning capabilities, along with concerning vulnerabilities in safety evaluations, particularly in high-risk domains such as biosecurity and chemical weapons. By open-sourcing our framework, we provide a foundation for developing more trustworthy AI systems and advancing research on model safety and ethics in scientific contexts.

cs.AI↗

arXiv2025

OmniFed: A Modular Framework for Configurable Federated Learning from Edge to HPC

Federated Learning (FL) is critical for edge and High Performance Computing (HPC) where data is not centralized and privacy is crucial. We present OmniFed, a modular framework designed around decoupling and clear separation of concerns for configuration, orchestration, communication, and training logic. Its architecture supports configuration-driven prototyping and code-level override-what-you-need customization. We also support different topologies, mixed communication protocols within a single deployment, and popular training algorithms. It also offers optional privacy mechanisms including Differential Privacy (DP), Homomorphic Encryption (HE), and Secure Aggregation (SA), as well as compression strategies. These capabilities are exposed through well-defined extension points, allowing users to customize topology and orchestration, learning logic, and privacy/compression plugins, all while preserving the integrity of the core system. We evaluate multiple models and algorithms to measure various performance metrics. By unifying topology configuration, mixed-protocol communication, and pluggable modules in one stack, OmniFed streamlines FL deployment across heterogeneous environments. Github repository is available at https://github.com/at-aaims/OmniFed.

cs.LG↗

arXiv2025

The (R)evolution of Scientific Workflows in the Agentic AI Era: Towards Autonomous Science

Modern scientific discovery increasingly requires coordinating distributed facilities and heterogeneous resources, forcing researchers to act as manual workflow coordinators rather than scientists. Advances in AI leading to AI agents show exciting new opportunities that can accelerate scientific discovery by providing intelligence as a component in the ecosystem. However, it is unclear how this new capability would materialize and integrate in the real world. To address this, we propose a conceptual framework where workflows evolve along two dimensions which are intelligence (from static to intelligent) and composition (from single to swarm) to chart an evolutionary path from current workflow management systems to fully autonomous, distributed scientific laboratories. With these trajectories in mind, we present an architectural blueprint that can help the community take the next steps towards harnessing the opportunities in autonomous science with the potential for 100x discovery acceleration and transformational scientific workflows.

cs.AI↗

arXiv2025

Scaling Up Data Parallelism in Decentralized Deep Learning

Although it has been extensively explored in theory, decentralized learning is not yet green-lighted for production use, largely due to a lack of stability, scalability, and generality in large scale DNN training. To shed light on the production use of decentralized learning, this work studies decentralized data parallel training at scale. To this end, we introduce a benchmarking framework, namely DBench, to host both centralized and decentralized DNN training. Building upon DBench, we introduce a benchmarking methodology to uncover the correlations between model accuracy and the variances of parameter tensors by varying communication graphs and training scales. Based on the benchmarking results, we observe that, (1) Similar to centralized learning, decentralized data parallel training also presents the issues of scalability and generality when the training scales up; (2) The model accuracy of decentralized learning is correlated to the number of connections in a communication graph; (3) The model accuracy of decentralized learning is surprisingly sensitive to the variance of parameter tensors across model replicas. Built upon the observations, we propose Ada, a decentralized adaptive approach that performs large scale DNN training following a decentralized SGD method and adapting the communication graph in use dynamically throughout training iterations. We apply Ada on large scale training and observe that Ada can obtain the best convergence rates consistently in decentralized DNN training, and delivers equally or comparably good model accuracy for all sample applications as centralized learning does, even when training ResNet50 for ImageNet-1K on the scale of 1008 GPUs.

cs.LG↗

arXiv2025

EPIC: Generative AI Platform for Accelerating HPC Operational Data Analytics

We present EPIC, an AI-driven platform designed to augment operational data analytics. EPIC employs a hierarchical multi-agent architecture where a top-level large language model provides query processing, reasoning and synthesis capabilities. These capabilities orchestrate three specialized low-level agents for information retrieval, descriptive analytics, and predictive analytics. This architecture enables EPIC to perform HPC operational analytics on multi-modal data, including text, images, and tabular formats, dynamically and iteratively. EPIC addresses the limitations of existing HPC operational analytics approaches, which rely on static methods that struggle to adapt to evolving analytics tasks and stakeholder demands. Through extensive evaluations on the Frontier HPC system, we demonstrate that EPIC effectively handles complex queries. Using descriptive analytics as a use case, fine-tuned smaller models outperform large state-of-the-art foundation models, achieving up to 26% higher accuracy. Additionally, we achieved 19x savings in LLM operational costs compared to proprietary solutions by employing a hybrid approach that combines large foundational models with fine-tuned local open-weight models.

cs.DB↗

arXiv2025

Decoding Memories: An Efficient Pipeline for Self-Consistency Hallucination Detection

Large language models (LLMs) have demonstrated impressive performance in both research and real-world applications, but they still struggle with hallucination. Existing hallucination detection methods often perform poorly on sentence-level generation or rely heavily on domain-specific knowledge. While self-consistency approaches help address these limitations, they incur high computational costs due to repeated generation. In this paper, we conduct the first study on identifying redundancy in self-consistency methods, manifested as shared prefix tokens across generations, and observe that non-exact-answer tokens contribute minimally to the semantic content. Based on these insights, we propose a novel Decoding Memory Pipeline (DMP) that accelerates generation through selective inference and annealed decoding. Being orthogonal to the model, dataset, decoding strategy, and self-consistency baseline, our DMP consistently improves the efficiency of multi-response generation and holds promise for extension to alignment and reasoning tasks. Extensive experiments show that our method achieves up to a 3x speedup without sacrificing AUROC performance.

cs.CL↗

arXiv2025

HPC Digital Twins for Evaluating Scheduling Policies, Incentive Structures and their Impact on Power and Cooling

Schedulers are critical for optimal resource utilization in high-performance computing. Traditional methods to evaluate schedulers are limited to post-deployment analysis, or simulators, which do not model associated infrastructure. In this work, we present the first-of-its-kind integration of scheduling and digital twins in HPC. This enables what-if studies to understand the impact of parameter configurations and scheduling decisions on the physical assets, even before deployment, or regarching changes not easily realizable in production. We (1) provide the first digital twin framework extended with scheduling capabilities, (2) integrate various top-tier HPC systems given their publicly available datasets, (3) implement extensions to integrate external scheduling simulators. Finally, we show how to (4) implement and evaluate incentive structures, as-well-as (5) evaluate machine learning based scheduling, in such novel digital-twin based meta-framework to prototype scheduling. Our work enables what-if scenarios of HPC systems to evaluate sustainability, and the impact on the simulated system.

cs.DC↗

arXiv2025

REALM: Recursive Relevance Modeling for LLM-based Document Re-Ranking

Large Language Models (LLMs) have shown strong capabilities in document re-ranking, a key component in modern Information Retrieval (IR) systems. However, existing LLM-based approaches face notable limitations, including ranking uncertainty, unstable top-k recovery, and high token cost due to token-intensive prompting. To effectively address these limitations, we propose REALM, an uncertainty-aware re-ranking framework that models LLM-derived relevance as Gaussian distributions and refines them through recursive Bayesian updates. By explicitly capturing uncertainty and minimizing redundant queries, REALM achieves better rankings more efficiently. Experimental results demonstrate that our REALM surpasses state-of-the-art re-rankers while significantly reducing token usage and latency, improving NDCG@10 by 0.7-11.9 and simultaneously reducing the number of LLM inferences by 23.4-84.4%, promoting it as the next-generation re-ranker for modern IR systems.

cs.IR↗

arXiv2025

X-MoE: Enabling Scalable Training for Emerging Mixture-of-Experts Architectures on HPC Platforms

Emerging expert-specialized Mixture-of-Experts (MoE) architectures, such as DeepSeek-MoE, deliver strong model quality through fine-grained expert segmentation and large top-k routing. However, their scalability is limited by substantial activation memory overhead and costly all-to-all communication. Furthermore, current MoE training systems - primarily optimized for NVIDIA GPUs - perform suboptimally on non-NVIDIA platforms, leaving significant computational potential untapped. In this work, we present X-MoE, a novel MoE training system designed to deliver scalable training performance for next-generation MoE architectures. X-MoE achieves this via several novel techniques, including efficient padding-free MoE training with cross-platform kernels, redundancy-bypassing dispatch, and hybrid parallelism with sequence-sharded MoE blocks. Our evaluation on the Frontier supercomputer, powered by AMD MI250X GPUs, shows that X-MoE scales DeepSeek-style MoEs up to 545 billion parameters across 1024 GPUs - 10x larger than the largest trainable model with existing methods under the same hardware budget, while maintaining high training throughput. The source code of X-MoE is available at https://github.com/Supercomputing-System-AI-Lab/X-MoE.

cs.LG↗

arXiv2025

Intelligent Sampling of Extreme-Scale Turbulence Datasets for Accurate and Efficient Spatiotemporal Model Training

With the end of Moore's law and Dennard scaling, efficient training increasingly requires rethinking data volume. Can we train better models with significantly less data via intelligent subsampling? To explore this, we develop SICKLE, a sparse intelligent curation framework for efficient learning, featuring a novel maximum entropy (MaxEnt) sampling approach, scalable training, and energy benchmarking. We compare MaxEnt with random and phase-space sampling on large direct numerical simulation (DNS) datasets of turbulence. Evaluating SICKLE at scale on Frontier, we show that subsampling as a preprocessing step can, in many cases, improve model accuracy and substantially lower energy consumption, with observed reductions of up to 38x.

cs.LG↗

arXiv2025

Data Readiness for Scientific AI at Scale

This paper examines how Data Readiness for AI (DRAI) principles apply to leadership-scale scientific datasets used to train foundation models. We analyze archetypal workflows across four representative domains - climate, nuclear fusion, bio/health, and materials - to identify common preprocessing patterns and domain-specific constraints. We introduce a two-dimensional readiness framework composed of Data Readiness Levels (raw to AI-ready) and Data Processing Stages (ingest to shard), both tailored to high performance computing (HPC) environments. This framework outlines key challenges in transforming scientific data for scalable AI training, emphasizing transformer-based generative models. Together, these dimensions form a conceptual maturity matrix that characterizes scientific data readiness and guides infrastructure development toward standardized, cross-domain support for scalable and reproducible AI for science.

cs.AI↗