arXiv subjects

Xuanyu Wang

Publications and source records attributed to Xuanyu Wang.

13 recordsLinked to original sources

arXiv2026

DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

Speculative decoding accelerates Large Language Model (LLM) inference by decoupling draft generation from target verification. While recent parallel drafters efficiently propose long token sequences in a single forward pass, they suffer from rapid acceptance decay due to a lack of inter-token dependencies. Furthermore, indiscriminately verifying these extended blocks wastes critical batch capacity on tokens with high rejection risks, severely degrading throughput in high-concurrency serving systems. We introduce DSpark, a speculative decoding framework that unifies high-throughput parallel generation with adaptive, load-aware verification. To maintain draft quality, DSpark utilizes a semi-autoregressive architecture, coupling a parallel backbone with a lightweight sequential module, to introduce intra-block dependency modeling and mitigate suffix decay. To optimize system efficiency, DSpark employs confidence-scheduled verification, dynamically tailoring the verification length for each request based on estimated prefix survival probabilities and engine-specific throughput profiles. On offline benchmarks across diverse domains, DSpark substantially improves the accepted length over state-of-the-art autoregressive and parallel drafters. When deployed within the DeepSeek-V4 serving system under live user traffic, DSpark successfully mitigates verification waste. Compared to the established production baseline (MTP-1), DSpark accelerates per-user generation speeds by 60 to 85 percent at matched throughput levels. More importantly, by preventing severe throughput degradation under strict interactivity constraints, it enables performance tiers that were previously unattainable, shifting the Pareto frontier of our serving system.

cs.AI↗

arXiv2026

Hetu v2: A General and Scalable Deep Learning System with Hierarchical and Heterogeneous Single Program Multiple Data Annotations

The Single-Program Multiple-Data (SPMD) paradigm provides a unified abstraction to annotate various parallel dimensions in distributed deep learning (DL) training. With SPMD, users can write training programs from the viewpoint of a single device, and the system will automatically deduce the tensor sharding and communication patterns. However, with the recent development in large-scale DL models, distributed training exhibits spatial and temporal workload heterogeneity, arising from both device disparities (e.g., mixed hardware, failures) and data variations (e.g., uneven sequence lengths). Such heterogeneity violates SPMD's assumption of symmetric workload partitioning, which restricts its ability to express and optimize heterogeneous parallel strategies effectively. To address this, we propose HSPMD within the Hetu v2 system to achieve general and scalable DL training. HSPMD extends SPMD's declarative annotations to support asymmetric sharding and composes standard communication primitives for hierarchical communication, all while retaining the simplicity of a single-device programming model. HSPMD handles spatial heterogeneity through progressive graph specialization, enabling device-specific execution logic, and addresses temporal heterogeneity via dynamic graph switching. Evaluations on (a) heterogeneous devices, (b) unstable devices, and (c) mixed-length data scenarios show that HSPMD matches or outperforms specialized systems, providing a flexible and efficient solution for modern distributed DL training.

cs.DC↗

arXiv2026

AcademiClaw: When Students Set Challenges for AI Agents

Benchmarks within the OpenClaw ecosystem have thus far evaluated exclusively assistant-level tasks, leaving the academic-level capabilities of OpenClaw largely unexamined. We introduce AcademiClaw, a bilingual benchmark of 80 complex, long-horizon tasks sourced directly from university students' real academic workflows -- homework, research projects, competitions, and personal projects -- that they found current AI agents unable to solve effectively. Curated from 230 student-submitted candidates through rigorous expert review, the final task set spans 25+ professional domains, ranging from olympiad-level mathematics and linguistics problems to GPU-intensive reinforcement learning and full-stack system debugging, with 16 tasks requiring CUDA GPU execution. Each task executes in an isolated Docker sandbox and is scored on task completion by multi-dimensional rubrics combining six complementary techniques, with an independent five-category safety audit providing additional behavioral analysis. Experiments on six frontier models show that even the best achieves only a 55\% pass rate. Further analysis uncovers sharp capability boundaries across task domains, divergent behavioral strategies among models, and a disconnect between token consumption and output quality, providing fine-grained diagnostic signals beyond what aggregate metrics reveal. We hope that AcademiClaw and its open-sourced data and code can serve as a useful resource for the OpenClaw community, driving progress toward agents that are more capable and versatile across the full breadth of real-world academic demands. All data and code are available at https://github.com/GAIR-NLP/AcademiClaw.

cs.AI↗

arXiv2026

A Python/CuPy Software Correlator for QUEST: Real-Time Performance and Initial Imaging

We present a Python/CuPy FX software correlator for small radio interferometer arrays and evaluate it on QUEST (Qilu University Explorer Survey Telescope). The system combines multi-threaded data ingest, pinned-memory host-device transfers, GPU-accelerated correlation, Polyphase Filter Bank channelization, MAD-based RFI flagging, and delay/phase calibration in a single workflow aimed at array commissioning. On a single NVIDIA RTX 4090D GPU, the implementation reaches a peak throughput of 1.51 GB/s, which is sufficient for real-time operation in the four-antenna mode tested here. After calibration, the visibility phase across a clean 1.32-1.38 GHz band is flattened to a residual scatter of a few degrees. Using the calibrated visibilities, we form a four-antenna synthesis image of Cassiopeia A; the CLEANed image recovers a compact source at the phase center and reduces image-domain background fluctuations from order 0.1 to a few 0.01 Jy/beam. These results indicate that the software is suitable for small-array commissioning and initial synthesis imaging on QUEST. A GNSS-based beam measurement is included as a supporting commissioning check.

astro-ph.IM↗

arXiv2026

RPIQ: Residual-Projected Multi-Collaboration Closed-Loop and Single Instance Quantization for Visually Impaired Assistance

Visually impaired users face significant challenges in daily information access and real-time environmental perception, and there is an urgent need for intelligent assistive systems with accurate recognition capabilities. Although large-scale models provide effective solutions for perception and reasoning, their practical deployment on assistive devices is severely constrained by excessive memory consumption and high inference costs. Moreover, existing quantization strategies often ignore inter-block error accumulation, leading to degraded model stability. To address these challenges, this study proposes a novel quantization framework -- Residual-Projected Multi-Collaboration Closed-Loop and Single Instance Quantization(RPIQ), whose quantization process adopts a multi-collaborative closed-loop compensation scheme based on Single Instance Calibration and Gauss-Seidel Iterative Quantization. Experiments on various types of large-scale models, including language models such as OPT, Qwen, and LLaMA, as well as vision-language models such as CogVLM2, demonstrate that RPIQ can compress models to 4-bit representation while significantly reducing peak memory consumption (approximately 60%-75% reduction compared to original full-precision models). The method maintains performance highly close to full-precision models across multiple language and visual tasks, and exhibits excellent recognition and reasoning capabilities in key applications such as text understanding and visual question answering in complex scenarios. While verifying the effectiveness of RPIQ for deployment in real assistive systems, this study also advances the computational efficiency and reliability of large models, enabling them to provide visually impaired users with the required information accurately and rapidly.

cs.LG↗

arXiv2026

Scene-Aware Vectorized Memory Multi-Agent Framework with Cross-Modal Differentiated Quantization VLMs for Visually Impaired Assistance

Visually impaired individuals face significant challenges in environmental perception. Traditional assistive technologies often lack adaptive intelligence, focusing on individual components rather than integrated systems. While Vision-Language Models (VLMs) offer a promising path to richer, integrated understanding, their deployment is severely limited by substantial computational requirements, demanding dozens of gigabytes of memory. To address these gaps in computational efficiency and integrated design, this study proposes a dual technological innovation framework: a cross-modal differentiated quantization framework for VLMs and a scene-aware vectorized memory multi-agent system. The quantization framework implements differentiated strategies, reducing memory from 38GB to 11.3GB. The multi-agent system uses vectorized memory and perception-memory-reasoning workflows to provide environmental information beyond the current view, achieving 2.83-3.52s latency to initial speech output. Experiments show the quantized 19B-parameter model only experiences a 2.05% performance drop on MMBench and maintains 63.7 accuracy on OCR-VQA (original: 64.9), outperforming smaller models with equivalent memory. This research advances computational efficiency and assistive technology, offering comprehensive assistance in scene perception, text recognition, and navigation.

cs.CV↗

arXiv2025

Deep Pathomic Learning Defines Prognostic Subtypes and Molecular Drivers in Colorectal Cancer

Precise prognostic stratification of colorectal cancer (CRC) remains a major clinical challenge due to its high heterogeneity. The conventional TNM staging system is inadequate for personalized medicine. We aimed to develop and validate a novel multiple instance learning model TDAM-CRC using histopathological whole-slide images for accurate prognostic prediction and to uncover its underlying molecular mechanisms. We trained the model on the TCGA discovery cohort (n=581), validated it in an independent external cohort (n=1031), and further we integrated multi-omics data to improve model interpretability and identify novel prognostic biomarkers. The results demonstrated that the TDAM-CRC achieved robust risk stratification in both cohorts. Its predictive performance significantly outperformed the conventional clinical staging system and multiple state-of-the-art models. The TDAM-CRC risk score was confirmed as an independent prognostic factor in multivariable analysis. Multi-omics analysis revealed that the high-risk subtype is closely associated with metabolic reprogramming and an immunosuppressive tumor microenvironment. Through interaction network analysis, we identified and validated Mitochondrial Ribosomal Protein L37 (MRPL37) as a key hub gene linking deep pathomic features to clinical prognosis. We found that high expression of MRPL37, driven by promoter hypomethylation, serves as an independent biomarker of favorable prognosis. Finally, we constructed a nomogram incorporating the TDAM-CRC risk score and clinical factors to provide a precise and interpretable clinical decision-making tool for CRC patients. Our AI-driven pathological model TDAM-CRC provides a robust tool for improved CRC risk stratification, reveals new molecular targets, and facilitates personalized clinical decision-making.

cs.LG↗

arXiv2025

ASTROFLOW: A Real-Time End-to-End Pipeline for Radio Single-Pulse Searches

Fast radio bursts (FRBs) are extremely bright, millisecond duration cosmic transients of unknown origin. The growing number of wide-field and high-time-resolution radio surveys, particularly with next-generation facilities such as the SKA and MeerKAT, will dramatically increase FRB discovery rates, but also produce data volumes that overwhelm conventional search pipelines. Real-time detection thus demands software that is both algorithmically robust and computationally efficient. We present Astroflow, an end-to-end, GPU-accelerated pipeline for single-pulse detection in radio time-frequency data. Built on a unified C++/CUDA core with a Python interface, Astroflow integrates RFI excision, incoherent dedispersion, dynamic-spectrum tiling, and a YOLO-based deep detector. Through vectorized memory access, shared-memory tiling, and OpenMP parallelism, it achieves 10x faster-than-real-time processing on consumer GPUs for a typical 150 s, 2048-channel observation, while preserving high sensitivity across a wide range of pulse widths and dispersion measures. These results establish the feasibility of a fully integrated, GPU-accelerated single-pulse search stack, capable of scaling to the data volumes expected from upcoming large-scale surveys. Astroflow offers a reusable and deployable solution for real-time transient discovery, and provides a framework that can be continuously refined with new data and models.

astro-ph.IM↗

arXiv2025

Hydraulis: Balancing Large Transformer Model Training via Co-designing Parallel Strategies and Data Assignment

To optimize large Transformer model training, both efficient parallel computing and advanced data management are indispensable. However, current methods often assume a stable and uniform training workload, neglecting data-induced imbalances-arising from both sampling and packing processes-which can impede training performance. Specifically, data sampling imbalance arises from uneven sequence length distribution of the training data, while data packing imbalance stems from the discrepancy between the linear memory complexity and quadratic time complexity of the attention mechanism. To address these imbalance issues, we develop Hydraulis, which jointly optimizes the parallel strategies and data assignment. For one thing, we introduce large model training with dynamic heterogeneous parallel strategies in response to the sequence length variations within and across training iterations. For another, we devise a two-stage data assignment approach, which strikes a good balance in terms of the training workloads both within and across model replicas. Empirical results demonstrate that Hydraulis outperforms existing systems by 1.32-2.66 times.

cs.DC↗

arXiv2025

LobRA: Multi-tenant Fine-tuning over Heterogeneous Data

With the breakthrough of Transformer-based pre-trained models, the demand for fine-tuning (FT) to adapt the base pre-trained models to downstream applications continues to grow, so it is essential for service providers to reduce the cost of processing FT requests. Low-rank adaption (LoRA) is a widely used FT technique that only trains small-scale adapters and keeps the base model unaltered, conveying the possibility of processing multiple FT tasks by jointly training different LoRA adapters with a shared base model. Nevertheless, through in-depth analysis, we reveal the efficiency of joint FT is dampened by two heterogeneity issues in the training data -- the sequence length variation and skewness. To tackle these issues, we develop LobRA, a brand new framework that supports processing multiple FT tasks by jointly training LoRA adapters. Two innovative designs are introduced. Firstly, LobRA deploys the FT replicas (i.e., model replicas for FT) with heterogeneous resource usages and parallel configurations, matching the diverse workloads caused by the sequence length variation. Secondly, for each training step, LobRA takes account of the sequence length skewness and dispatches the training data among the heterogeneous FT replicas to achieve workload balance. We conduct experiments to assess the performance of LobRA, validating that it significantly reduces the GPU seconds required for joint FT by 45.03%-60.67%.

cs.DC↗

arXiv2025

Subspecialty-Specific Foundation Model for Intelligent Gastrointestinal Pathology

Gastrointestinal (GI) diseases represent a clinically significant burden, necessitating precise diagnostic approaches to optimize patient outcomes. Conventional histopathological diagnosis suffers from limited reproducibility and diagnostic variability. To overcome these limitations, we develop Digepath, a specialized foundation model for GI pathology. Our framework introduces a dual-phase iterative optimization strategy combining pretraining with fine-screening, specifically designed to address the detection of sparsely distributed lesion areas in whole-slide images. Digepath is pretrained on over 353 million multi-scale images from 210,043 H&E-stained slides of GI diseases. It attains state-of-the-art performance on 33 out of 34 tasks related to GI pathology, including pathological diagnosis, protein expression status prediction, gene mutation prediction, and prognosis evaluation. We further translate the intelligent screening module for early GI cancer and achieve near-perfect 99.70% sensitivity across nine independent medical institutions. This work not only advances AI-driven precision pathology for GI diseases but also bridge critical gaps in histopathological practice.

eess.IV↗

arXiv2025

Malleus: Straggler-Resilient Hybrid Parallel Training of Large-scale Models via Malleable Data and Model Parallelization

As the scale of models and training data continues to grow, there is an expanding reliance on more GPUs to train large-scale models, which inevitably increases the likelihood of encountering dynamic stragglers that some devices lag behind in performance occasionally. However, hybrid parallel training, one of the de facto paradigms to train large models, is typically sensitive to the stragglers. This paper presents Malleus, a straggler-resilient hybrid parallel training framework for large-scale models. Malleus quantifies the stragglers at the nuanced, per-GPU granularity during training, and develops a novel planning algorithm to deduce the optimal parallelization of GPU devices, pipeline stages, model layers, and training data, maximizing training efficiency when stragglers exist. In addition, once a shift in the straggler situation is detected, Malleus adaptively adjusts the parallelization via a re-planning process, and seamlessly and efficiently migrates the model states on the fly, without sacrificing the stability of the training tasks. Empirical results on large language models with up to 110B parameters show that Malleus consistently outperforms existing parallel training frameworks under various straggler situations, delivering on average 2.63-5.28 times of efficiency improvement.

cs.DC↗

arXiv2024

Real-and-Present: Investigating the Use of Life-Size 2D Video Avatars in HMD-Based AR Teleconferencing

Augmented Reality (AR) teleconferencing allows separately located users to interact with each other in 3D through agents in their own physical environments. Existing methods leveraging volumetric capturing and reconstruction can provide a high-fidelity experience but are often too complex and expensive for everyday usage. Other solutions target mobile and effortless-to-setup teleconferencing on AR Head Mounted Displays (HMD). They directly transplant the conventional video conferencing onto an AR-HMD platform or use avatars to represent remote participants. However, they can only support either a high fidelity or a high level of co-presence. Moreover, the limited Field of View (FoV) of HMDs could further influence users' immersive experience. To achieve a balance between fidelity and co-presence, we explore using life-size 2D video-based avatars (video avatars for short) in AR teleconferencing. Specifically, with the potential effect of FoV on users' perception of proximity, we first conduct a pilot study to explore the local-user-centered optimal placement of video avatars in small-group AR conversations. With the placement results, we then implement a proof-of-concept prototype of video-avatar-based teleconferencing. We conduct user evaluations with the prototype to verify its effectiveness in balancing fidelity and co-presence. Following the indication in the pilot study, we further quantitatively explore the effect of FoV size on the video avatar's optimal placement through a user study involving more FoV conditions in a VR-simulated environment. We regress placement models to serve as references for computationally determining video avatar placements in such teleconferencing applications on various existing AR HMDs and future ones with bigger FoVs.

cs.HC↗