arXiv subjects

Hongli Xu

Publications and source records attributed to Hongli Xu.

At least 19 recordsLinked to original sources

arXiv2026

Pallas: A Proactive KV Cache Migration Framework for LLM Inference in AI-RAN

AI-RAN brings large language model (LLM) serving close to mobile users, but cellular handover can separate an active request from its inference state: the user attaches to a target base station (gNB) while the large and growing key-value (KV) cache remains at the source. Retaining inference at the source preserves service continuity but persistently increases inter-token latency (ITL), whereas recovering the state at the target restores serving locality but requires KV-cache transfer, recomputation, or a combination of both only after handover, directly prolonging service interruption time (SIT). This work presents Pallas, a \textit{proactive} KV-cache migration framework that prepares the inference state at the predicted target before handover, in parallel with ongoing source-side inference and token delivery. At the preparation trigger, Pallas partitions the token sequence into a stable historical prefix and an evolving suffix. The target reconstructs the prefix through local prefill, while the source streams the KV blocks generated for the suffix. At handover, the target assembles both portions into an up-to-date KV cache and resumes decoding locally, leaving only unfinished preparation to contribute to SIT. An online scheduler selects the \textit{prefetching window}, which determines how early preparation begins before handover, based on mobility predictions and runtime telemetry. Across three LLMs and $100$--$500~\mathrm{Mbps}$ inter-gNB links, our vLLM-based prototype reduces average SIT by factors of $2.28$--$89.68$ over target-side recovery approaches and lowers average ITL by $16.0\%$--$50.0\%$ compared with source-side forwarding.

cs.LG↗

arXiv2026

SOS! : A Streamlined Object-Conditional Transformer for Model-free Segmentation

Foundation segmentation models excel at generating high-quality, class-agnostic masks, but they struggle to associate these proposals with specific target objects. This semantic gap severely hinders their deployment in downstream applications like robotic manipulation, which demand precise unseen objects segmentation. Existing approaches attempt to resolve this by relying on exhaustive 3D object model priors, inherently introducing prohibitive computational overhead and complex, multi-stage pipelines. To address these limitations, we propose SOS (Streamlined Object-conditional Transformer for model-free Segmentation). SOS completely eliminates the reliance on 3D models, requiring only a single reference image per target object. Central to our framework is a novel Object-Conditional Transformer that learns identity-anchored queries, unifying mask generation and target identification into a single feed-forward pass. This streamlined design drastically improves both structural and computational efficiency. Extensive evaluations across multiple benchmarks demonstrate that SOS establishes a new state-of-the-art for model-free unseen objects segmentation, delivering accurate and high-efficiency performance. The project page and code are available at https://sos-seg.github.io/.

cs.CV↗

arXiv2026

Pose Anything Anywhere:Model-free Object Poses from Arbitrary References

Estimating the 6D pose of unseen objects is a fundamental yet challenging problem for open-world robotics and embodied perception. Model-based methods are accurate but depend on CAD assets or heavy onboarding, while most model-free approaches are still limited to pairwise single-anchor matching and thus fail under occlusion and large viewpoint changes with low query-reference overlap. Therefore, we present PANY, a unified model-free framework that seamlessly supports both RGB and RGB-D inputs, operates on one or sparse pose-free reference views, and generalizes effectively to novel objects. Built on a multi-view transformer geometry backbone, PANY moves beyond pairwise matching by learning view-consistent geometry and cross-view alignment cues that remain stable under wide baselines and limited overlap. When additional unposed assist views are available, PANY aggregates them via pose-graph canonical registration to increase geometric coverage and reinforce the final pose. Extensive experiments show that PANY achieves state-of-the-art performance across multiple benchmarks, substantially outperforming existing model-free methods, improving pose accuracy by +12% on YCB-V and over +20% on LM-O. Furthermore, PANY consistently performs well under both single-reference and sparse-reference settings, demonstrating strong robustness in real-world environments.

cs.CV↗

arXiv2026

Bandwidth-Aware and Cost-Efficient Pipeline Parallel Scheduling in Geo-Distributed LLM Training

The rapid evolution of large language models (LLMs) has made geographically distributed training necessary due to GPU scarcity within a single cloud region. In such cross-region settings, Pipeline Parallelism (PP) is communication-efficient, yet scheduling PP remains challenging under heterogeneous inter-region bandwidth and regional electricity prices. Existing schedulers are either delay-first, incurring high electricity cost, or cost-first, relying on rigid resource allocation that prolongs Job Completion Time (JCT). They are also ineffective at optimizing execution order in multi-tenant environments, where long-running and bandwidth-intensive jobs can cause head-of-line (HoL) blocking and degrade overall performance. To this end, we propose BACE-Pipe, a bandwidth-aware and cost-efficient pipeline scheduling framework for LLM training across geo-distributed clusters. BACE-Pipe first introduces a dynamic job prioritization mechanism that optimizes execution order by jointly considering job characteristics (e.g., computation time) and real-time network utilization. It then employs a bandwidth-aware pathfinder to identify feasible cross-region pipeline paths that satisfy communication constraints, thereby preventing communication from stalling the pipeline. Among all feasible paths, a cost-minimizing allocator determines the optimal GPU placement strategy by preferentially assigning resources to regions with lower electricity prices. Consequently, BACE-Pipe mitigates HoL blocking, improves resource utilization, and simultaneously reduces both JCT and total electricity cost. Extensive simulations show that BACE-Pipe reduces average JCT by 27.9%--64.7% and total electricity cost by 12.6%--30.6% compared with state-of-the-art baselines.

cs.DC↗

arXiv2026

ReLoRA: Knowledge-Reusing Adaptation for Fast Rollout of Evolving LLM Services

Large Language Models (LLMs) are increasingly deployed as continuously evolving services, where frequent base-model updates may invalidate previously deployed task-specific Low-Rank Adaptation (LoRA) adapters. For service providers managing numerous downstream model services, retraining each LoRA adapter from scratch for every updated base model is computationally prohibitive and delays service rollout. Meanwhile, the simpler alternative, i.e., naively applying the original LoRA adapter to the updated base model, often leads to degraded service quality due to adapter-backbone incompatibility. To address this problem, we propose ReLoRA, a knowledge-reusing re-adaptation framework that efficiently restores service-ready LoRA adapters for evolving LLM services while preserving or improving task performance. Specifically, ReLoRA comprises two key optimization steps: 1) Adaptive LoRA initialization leverages Bayesian optimization to construct a compatibility-aware starting point by fusing information from both the previously deployed task adapter and the base model's evolution; 2) Fine-tuning with scheduled regularization first rapidly steers the adapter to a high-quality region via strong regularization, followed by relaxed regularization for task-specific refinement. This design enables rapid service-quality recovery with reduced re-adaptation overhead. Extensive experiments demonstrate that ReLoRA reduces time-to-readiness by up to 8.9$\times$ and improves accuracy by up to 4.6\% compared to baselines.

cs.LG↗

arXiv2026

Adaptive and Fine-grained Module-wise Expert Pruning for Efficient LoRA-MoE Fine-Tuning

LoRA-MoE has emerged as an effective paradigm for parameter-efficient fine-tuning, combining the low training cost of LoRA with the increased adaptation capacity of Mixture-of-Experts (MoE). However, existing LoRA-MoE frameworks typically adopt a fixed and uniform expert configuration across heterogeneous Transformer modules (\eg, attention query/key projections and MLP gating networks), ignoring their distinct functional roles and capacity requirements. This design leads to localized over-provisioning, redundant trainable parameters, and unnecessary optimizer-state overhead. Moreover, prior methods enforce load balancing among experts throughout training. Although beneficial in the early stage, this constraint becomes restrictive once routing patterns stabilize, limiting expert specialization on downstream tasks. In this paper, we propose DMEP, a novel LoRA-MoE fine-tuning framework based on Dynamic Module-wise Expert Pruning. DMEP tracks expert utilization during training and physically removes low-utility experts on a per-module basis, yielding a more compact expert structure tailored to different modules. The pruned model then continues training without the load-balancing constraint, freeing the remaining experts to focus entirely on the downstream task and develop specialized expertise. By jointly adapting module-wise expert capacity and eliminating unnecessary balancing, DMEP improves both parameter efficiency and training efficiency. Extensive experiments on multiple reasoning benchmarks show that DMEP reduces trainable parameters by 35\%--43\% and improves training throughput by about 10\%, while maintaining or surpassing the downstream reasoning accuracy of uniform LoRA-MoE baselines.

cs.LG↗

arXiv2026

Less is More: Lightweight Prompt Compression for Question Answering Applications on Edge Devices

In agent-driven question answering (QA) applications, retrieval-augmented generation (RAG) is commonly introduced to enhance the response accuracy of large language models (LLMs) by providing additional context. Due to the inherent noise in retrieval results and the coarse granularity of document-level retrieval, the retrieved context often contains substantial redundant information. In this setting, the agent prompt, consisting of the user query and the associated retrieved context, leads to unnecessary computational overhead during LLM inference. Existing prompt compression methods typically rely on auxiliary small language models (SLMs) to estimate context importance. However, such approaches introduce significant memory and computational overhead, which limits their deployment on resource-constrained edge devices. In this paper, we propose CORE, a two-stage sentence-level prompt compression method that eliminates the need for SLMs. In the first stage, CORE constructs an answer set via named entity recognition (NER) and a clue set via semantic matching. In the second stage, CORE refines the clue set using an orthogonal residual retrieval strategy and designs a spatial proximity-based metric to filter the answer set. The two sets are then combined to form the final compressed context. We implement CORE on an NVIDIA Jetson AGX Orin edge device and a Huawei Nova smartphone. Experimental results demonstrate that within a 2000-token budget, CORE improves accuracy by at least 30.19% compared to state-of-the-art baselines, while reducing memory usage by at least 50.47% and achieving at least 1.94 times speedup on the edge device. Moreover, compared to the state-of-the-art LLMLingua2 method, CORE achieves a substantial energy reduction of 95.74% on the smartphone, highlighting its practicality and generalizability for mobile deployments.

cs.CL↗

arXiv2026

Utility-Guided Agent Orchestration for Efficient LLM Tool Use

Tool-using large language model (LLM) agents often face a fundamental tension between answer quality and execution cost. Fixed workflows are stable but inflexible, while free-form multi-step reasoning methods such as ReAct may improve task performance at the expense of excessive tool calls, longer trajectories, higher token consumption, and increased latency. In this paper, we study agent orchestration as an explicit decision problem rather than leaving it entirely to prompt-level behavior. We propose a utility-guided orchestration policy that selects among actions such as respond, retrieve, tool call, verify, and stop by balancing estimated gain, step cost, uncertainty, and redundancy. Our goal is not to claim universally best task performance, but to provide a controllable and analyzable policy framework for studying quality-cost trade-offs in tool-using LLM agents. Experiments across direct answering, threshold control, fixed workflows, ReAct, and several policy variants show that explicit orchestration signals substantially affect agent behavior. Additional analyses on cost definitions, workflow fairness, and redundancy control further demonstrate that lightweight utility design can provide a defensible and practical mechanism for agent control.

cs.AI↗

arXiv2026

Xray-Visual Models: Scaling Vision models on Industry Scale Data

We present Xray-Visual, a unified vision model architecture for large-scale image and video understanding trained on industry-scale social media data. Our model leverages over 15 billion curated image-text pairs and 10 billion video-hashtag pairs from Facebook and Instagram, employing robust data curation pipelines that incorporate balancing and noise suppression strategies to maximize semantic diversity while minimizing label noise. We introduce a three-stage training pipeline that combines self-supervised MAE, semi-supervised hashtag classification, and CLIP-style contrastive learning to jointly optimize image and video modalities. Our architecture builds on a Vision Transformer backbone enhanced with efficient token reorganization (EViT) for improved computational efficiency. Extensive experiments demonstrate that Xray-Visual achieves state-of-the-art performance across diverse benchmarks, including ImageNet for image classification, Kinetics and HMDB51 for video understanding, and MSCOCO for cross-modal retrieval. The model exhibits strong robustness to domain shift and adversarial perturbations. We further demonstrate that integrating large language models as text encoders (LLM2CLIP) significantly enhances retrieval performance and generalization capabilities, particularly in real-world environments. Xray-Visual establishes new benchmarks for scalable, multimodal vision models, while maintaining superior accuracy and computational efficiency.

cs.CV↗

arXiv2025

Improving LLM Reasoning via Dependency-Aware Query Decomposition and Logic-Parallel Content Expansion

The integration of Large Language Models (LLMs) into real-time Web applications, such as AI-powered search and conversational agents, presents a fundamental Web infrastructure challenge: reconciling the demand for high-quality, complex reasoning with the stringent low-latency and high-throughput requirements of interactive services. Current LLM reasoning, hindered by computationally inefficient sequential generation and rigid reasoning strategies, creates a critical bottleneck for the Web services. Existing approaches typically optimize the LLM reasoning for either efficiency or quality but struggle to achieve both, and thus fail to meet the dual requirements of modern Web platforms. To overcome these limitations, we propose Orion, a novel and efficient reasoning framework that enables dependency-aware query decomposition and logic-parallel content expansion. Concretely, Orion decomposes a single query reasoning process into two synergistic phases: (1) \textit{key point generation}, which distills logically structured key points through retrieval-augmented few-shot prompting, and (2) \textit{content parallel expansion}, which concurrently elaborates on these points based on a dependency graph to ensure logical consistency. Furthermore, Orion introduces a pipeline scheduling mechanism that exploits the complementary computational characteristics of the two phases (generation imposes pressure on GPU computing and expansion stresses on GPU memory) across multiple queries, enabling cross-query parallelism and dramatically improving reasoning performance (\ie, efficiency and quality). Experiments on diverse benchmarks show that Orion not only delivers up to 4.33x higher token generation speed and 3.42x lower answer latency over the baselines but also improves reasoning quality by up to 18.75% through explicitly modeling inter-point dependencies.

cs.AI↗

arXiv2025

SABlock: Semantic-Aware KV Cache Eviction with Adaptive Compression Block Size

The growing memory footprint of the Key-Value (KV) cache poses a severe scalability bottleneck for long-context Large Language Model (LLM) inference. While KV cache eviction has emerged as an effective solution by discarding less critical tokens, existing token-, block-, and sentence-level compression methods struggle to balance semantic coherence and memory efficiency. To this end, we introduce SABlock, a \underline{s}emantic-aware KV cache eviction framework with \underline{a}daptive \underline{block} sizes. Specifically, SABlock first performs semantic segmentation to align compression boundaries with linguistic structures, then applies segment-guided token scoring to refine token importance estimation. Finally, for each segment, a budget-driven search strategy adaptively determines the optimal block size that preserves semantic integrity while improving compression efficiency under a given cache budget. Extensive experiments on long-context benchmarks demonstrate that SABlock consistently outperforms state-of-the-art baselines under the same memory budgets. For instance, on Needle-in-a-Haystack (NIAH), SABlock achieves 99.9% retrieval accuracy with only 96 KV entries, nearly matching the performance of the full-cache baseline that retains up to 8K entries. Under a fixed cache budget of 1,024, SABlock further reduces peak memory usage by 46.28% and achieves up to 9.5x faster decoding on a 128K context length.

cs.CL↗

arXiv2025

BioDet: Boosting Industrial Object Detection with Image Preprocessing Strategies

Accurate 6D pose estimation is essential for robotic manipulation in industrial environments. Existing pipelines typically rely on off-the-shelf object detectors followed by cropping and pose refinement, but their performance degrades under challenging conditions such as clutter, poor lighting, and complex backgrounds, making detection the critical bottleneck. In this work, we introduce a standardized and plug-in pipeline for 2D detection of unseen objects in industrial settings. Based on current SOTA baselines, our approach reduces domain shift and background artifacts through low-light image enhancement and background removal guided by open-vocabulary detection with foundation models. This design suppresses the false positives prevalent in raw SAM outputs, yielding more reliable detections for downstream pose estimation. Extensive experiments on real-world industrial bin-picking benchmarks from BOP demonstrate that our method significantly boosts detection accuracy while incurring negligible inference overhead, showing the effectiveness and practicality of the proposed method.

cs.CV↗

arXiv2025

Enabling Reconfiguration-Communication Overlap for Collective Communication in Optical Networks

Collective communication (CC) is critical for scaling distributed machine learning (DML). The predictable traffic patterns of DML present a great opportunity for applying optical network technologies. Optical networks with reconfigurable topologies promise high bandwidth and low latency for collective communications. However, existing approaches face inherent limitations: static topologies are inefficient for dynamic communication patterns within CC algorithm, while frequent topology reconfiguration matching every step of the algorithm incurs significant overhead. In this paper, we propose SWOT, a demand-aware optical network framework that employs ``intra-collective reconfiguration'' to dynamically align network resources with CC traffic patterns. SWOT hides reconfiguration latency by overlapping it with data transmission through three key techniques: \textit{Heterogeneous Message Splitting}, \textit{Asynchronous Overlapping}, and \textit{Topology Bypassing}. Extensive simulations demonstrate that SWOT reduces communication completion time up to 89.7% across diverse CC algorithm compared to static baselines, demonstrating strong robustness to varying optical resources and reconfiguration delay.

cs.NI↗

arXiv2025

FUNCanon: Learning Pose-Aware Action Primitives via Functional Object Canonicalization for Generalizable Robotic Manipulation

General-purpose robotic skills from end-to-end demonstrations often leads to task-specific policies that fail to generalize beyond the training distribution. Therefore, we introduce FunCanon, a framework that converts long-horizon manipulation tasks into sequences of action chunks, each defined by an actor, verb, and object. These chunks focus policy learning on the actions themselves, rather than isolated tasks, enabling compositionality and reuse. To make policies pose-aware and category-general, we perform functional object canonicalization for functional alignment and automatic manipulation trajectory transfer, mapping objects into shared functional frames using affordance cues from large vision language models. An object centric and action centric diffusion policy FuncDiffuser trained on this aligned data naturally respects object affordances and poses, simplifying learning and improving generalization ability. Experiments on simulated and real-world benchmarks demonstrate category-level generalization, cross-task behavior reuse, and robust sim2real deployment, showing that functional canonicalization provides a strong inductive bias for scalable imitation learning in complex manipulation domains. Details of the demo and supplemental material are available on our project website https://sites.google.com/view/funcanon.

cs.RO↗

arXiv2025

Accelerating Mixture-of-Expert Inference with Adaptive Expert Split Mechanism

Mixture-of-Experts (MoE) has emerged as a promising architecture for modern large language models (LLMs). However, massive parameters impose heavy GPU memory (i.e., VRAM) demands, hindering the widespread adoption of MoE LLMs. Offloading the expert parameters to CPU RAM offers an effective way to alleviate the VRAM requirements for MoE inference. Existing approaches typically cache a small subset of experts in VRAM and dynamically prefetch experts from RAM during inference, leading to significant degradation in inference speed due to the poor cache hit rate and substantial expert loading latency. In this work, we propose MoEpic, an efficient MoE inference system with a novel expert split mechanism. Specifically, each expert is vertically divided into two segments: top and bottom. MoEpic caches the top segment of hot experts, so that more experts will be stored under the limited VRAM budget, thereby improving the cache hit rate. During each layer's inference, MoEpic predicts and prefetches the activated experts for the next layer. Since the top segments of cached experts are exempt from fetching, the loading time is reduced, which allows efficient transfer-computation overlap. Nevertheless, the performance of MoEpic critically depends on the cache configuration (i.e., each layer's VRAM budget and expert split ratio). To this end, we propose a divide-and-conquer algorithm based on fixed-point iteration for adaptive cache configuration. Extensive experiments on popular MoE LLMs demonstrate that MoEpic can save about half of the GPU cost, while lowering the inference latency by about 37.51%-65.73% compared to the baselines.

cs.LG↗

arXiv2025

Towards Communication-Efficient Decentralized Federated Graph Learning over Non-IID Data

Decentralized Federated Graph Learning (DFGL) overcomes potential bottlenecks of the parameter server in FGL by establishing a peer-to-peer (P2P) communication network among workers. However, while extensive cross-worker communication of graph node embeddings is crucial for DFGL training, it introduces substantial communication costs. Most existing works typically construct sparse network topologies or utilize graph neighbor sampling methods to alleviate the communication overhead in DFGL. Intuitively, integrating these methods may offer promise for doubly improving communication efficiency in DFGL. However, our preliminary experiments indicate that directly combining these methods leads to significant training performance degradation if they are jointly optimized. To address this issue, we propose Duplex, a unified framework that jointly optimizes network topology and graph sampling by accounting for their coupled relationship, thereby significantly reducing communication cost while enhancing training performance in DFGL. To overcome practical DFGL challenges, eg, statistical heterogeneity and dynamic network environments, Duplex introduces a learning-driven algorithm to adaptively determine optimal network topologies and graph sampling ratios for workers. Experimental results demonstrate that Duplex reduces completion time by 20.1%--48.8% and communication costs by 16.7%--37.6% to achieve target accuracy, while improving accuracy by 3.3%--7.9% under identical resource budgets compared to baselines.

cs.DC↗

arXiv2025

Adaptive KV-Cache Compression without Manually Setting Budget

Large language models (LLMs) inference relies heavily on KV-caches to accelerate autoregressive decoding, but the resulting memory footprint grows rapidly with sequence length, posing significant efficiency challenges. Current KV-cache compression methods suffer from a Procrustes' bed problem: they force diverse workloads into fixed compression ratios, leading to suboptimal resource allocation and inference performance. To this end, we present GVote, an adaptive KV-cache compression scheme that eliminates manual budget specification while achieving superior accuracy-efficiency trade-offs. GVote operates on the principle that the important keys are the aggregation of keys required by future queries. The method predicts future query attention demands by Monte-Carlo style sampling potential queries and aggregating selected keys to determine the optimal cache budget without manual specification. Experimental evaluation demonstrates GVote's effectiveness across multiple benchmarks, including GSM8K, RULER and Longbench. Compared to baselines, GVote exhibits 2$\times$ memory reduction while the accuracy maintains higher or comparable.

cs.DB↗

arXiv2025

DySTop

Federated Learning (FL) has emerged as a potential distributed learning paradigm that enables model training on edge devices (i.e., workers) while preserving data privacy. However, its reliance on a centralized server leads to limited scalability. Decentralized federated learning (DFL) eliminates the dependency on a centralized server by enabling peer-to-peer model exchange. Existing DFL mechanisms mainly employ synchronous communication, which may result in training inefficiencies under heterogeneous and dynamic edge environments. Although a few recent asynchronous DFL (ADFL) mechanisms have been proposed to address these issues, they typically yield stale model aggregation and frequent model transmission, leading to degraded training performance on non-IID data and high communication overhead. To overcome these issues, we present DySTop, an innovative mechanism that jointly optimizes dynamic staleness control and topology construction in ADFL. In each round, multiple workers are activated, and a subset of their neighbors is selected to transmit models for aggregation, followed by local training. We provide a rigorous convergence analysis for DySTop, theoretically revealing the quantitative relationships between the convergence bound and key factors such as maximum staleness, activating frequency, and data distribution among workers. From the insights of the analysis, we propose a worker activation algorithm (WAA) for staleness control and a phase-aware topology construction algorithm (PTCA) to reduce communication overhead and handle data non-IID. Extensive evaluations through both large-scale simulations and real-world testbed experiments demonstrate that our DySTop reduces completion time by 51.8% and the communication resource consumption by 57.1% compared to state-of-the-art solutions, while maintaining the same model accuracy.

cs.DC↗