Searcharxiv⌕ Search

arXiv subjects

Hongwei Sun

Publications and source records attributed to Hongwei Sun.

At least 19 recordsLinked to original sources
arXiv2026

StrataCL: Fabric-Native Communication Library for Production Supernodes

Modern distributed AI workloads run across hundreds of accelerators, making communication a major bottleneck. Existing communication libraries remain largely buffer-centric because user and communication buffers are managed separately, causing redundant data copies or costly user-buffer registration. This paper presents StrataCL, a zero-redundancy and fabric-native communication library for production supernodes. StrataCL introduces registration-on-allocation to realize user-buffer direct communication, and designs communication operators with workload-balanced NPU-core partitioning and NPU-driven SDMA offloading to exploit supernode architecture features. On the Huawei CloudMatrix384, StrataCL improves collective bus bandwidth by up to 1.6x and improves MoE dispatch/combine bus bandwidth by up to 1.4x. Across three production workloads, StrataCL improves LLM inference throughput by 1.9x, reduces P99 TTFT by 2.2x, and reduces LLM and Recsys training iteration time by 1.4x and 1.3x, respectively.

cs.DC↗
arXiv2026

Relay Buffer Independent Communication over Pooled HBM for Efficient MoE Inference on Ascend

Mixture-of-Experts (MoE) inference requires large-scale token exchange across devices, making dispatch and combine major bottlenecks in both prefill and decode. Beyond network transfer, routing-driven layout transformation, temporary relay, and output restoration can add substantial overhead. Existing MoE communication paths are often buffer-centric, using explicit inter-process relay and reordering buffers around collective transfer. This report presents a relay-buffer-free communication design for MoE inference acceleration on Ascend systems. The design reorganizes dispatch and combine around direct placement into destination expert windows and direct reading from remote expert windows. Built on globally pooled high-bandwidth memory and symmetric-memory allocation, it removes most intermediate relay and reordering buffers while retaining only lightweight control state, including counts, offsets, and synchronization metadata. We instantiate the design as two schedules for the main phases of MoE inference: a prefill schedule with richer planning state for throughput-oriented execution, and a compact decode schedule for latency-sensitive execution. Experiments on Ascend-based MoE workloads show reduced dispatch and combine latency in both settings. At the serving level, the implementation improves time to first token (TTFT), preserves competitive time per output token (TPOT), and enlarges the feasible scheduling space under practical latency constraints. These results indicate that, on platforms with globally addressable device memory, reducing intermediate buffering and output restoration around expert execution is an effective direction for accelerating MoE inference.

cs.DC↗
arXiv2026

Spatial and temporal distribution of stratospheric turbulence from global high-resolution radiosonde data

Stratospheric turbulence is difficult to observe, yet it strongly affects how momentum, trace gases, and aerosols spread through the atmosphere. Here, we use global high-resolution radiosonde observations from 370 stations during October 2014-December 2025 to estimate stratospheric turbulent diffusivity with a Richardson-number-based method. We find that turbulent diffusivity spans a broad range and is dominated by turbulence occurring in statically stable but strongly sheared conditions. Enhanced values appear over Turkey, India, Malaysia, Japan, and major mountainous regions, with evidence that mountain waves and convective activity both contribute. We also identify a local maximum just above the tropical tropopause, between the Equator and 15 N at 17 km altitude, which may indicate an injection region conducive to the rapid aerosol dispersion for stratospheric aerosol injection (SAI). Stratospheric turbulent diffusivity also shows a significant increase ($3.5 \times 10^{-3}\ \mathrm{m^2\ s^{-1}\ yr^{-1}}$) over 2015-2025. Finally, these results may help constrain and predict the dispersion of stratospheric plumes from rockets, aircraft, and potential SAI.

physics.ao-ph↗
arXiv2025

FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs

Large language models (LLMs) have made a profound impact across various fields due to their advanced capabilities. However, training these models at unprecedented scales requires extensive AI accelerator clusters and sophisticated parallelism strategies, which pose significant challenges in maintaining system reliability over prolonged training periods. A major concern is the substantial loss of training time caused by inevitable hardware and software failures. To address these challenges, we present FlashRecovery, a fast and low-cost failure recovery system comprising three core modules: (1) Active and real-time failure detection. This module performs continuous training state monitoring, enabling immediate identification of hardware and software failures within seconds, thus ensuring rapid incident response; (2) Scale-independent task restart. By employing different recovery strategies for normal and faulty nodes, combined with an optimized communication group reconstruction protocol, our approach ensures that the recovery time remains nearly constant, regardless of cluster scale; (3) Checkpoint-free recovery within one step. Our novel recovery mechanism enables single-step restoration, completely eliminating dependence on traditional checkpointing methods and their associated overhead. Collectively, these innovations enable FlashRecovery to achieve optimal Recovery Time Objective (RTO) and Recovery Point Objective (RPO), substantially improving the reliability and efficiency of long-duration LLM training. Experimental results demonstrate that FlashRecovery system can achieve training restoration on training cluster with 4, 800 devices in 150 seconds. We also verify that the time required for failure recovery is nearly consistent for different scales of training tasks.

cs.DC↗
arXiv2025

Cloud-aerosol interactions in subtropical marine stratocumulus weaken in a warmer climate

Radiative effects of aerosol-cloud interactions constitute the most uncertain climate forcing of the Earth system, making it important to understand how they may change with climate. We conduct 3-day-long large-eddy simulations of a stratocumulus-to-cumulus transition along an airmass-following trajectory over the Northeast Pacific Ocean. By perturbing boundary layer aerosol concentrations, we simulate aerosol-cloud interactions in both present-day and doubled-CO2 conditions. Aerosol-induced cloud changes, including the Twomey effect and adjustments of cloud fraction and liquid water path, are inhibited in a doubled-CO2 climate. Decomposing the aerosol-induced cloud radiative effect change ($\Delta$CRE) reveals that aerosol-induced cloud fraction changes dominate $\Delta$CRE. Overall, doubling CO2 attenuates aerosol-induced $\Delta$CRE (i.e., cooling) by >30% in our simulations. Our results also show that low cloud feedbacks are sensitive to the background aerosol concentration, highlighting the interplay between climate forcings and feedbacks. These results may aid in predicting the cooling potential of marine cloud brightening in a changing climate.

physics.ao-ph↗
arXiv2025

Fast and Accurate Power Load Data Completion via Regularization-optimized Low-Rank Factorization

Low-rank representation learning has emerged as a powerful tool for recovering missing values in power load data due to its ability to exploit the inherent low-dimensional structures of spatiotemporal measurements. Among various techniques, low-rank factorization models are favoured for their efficiency and interpretability. However, their performance is highly sensitive to the choice of regularization parameters, which are typically fixed or manually tuned, resulting in limited generalization capability or slow convergence in practical scenarios. In this paper, we propose a Regularization-optimized Low-Rank Factorization, which introduces a Proportional-Integral-Derivative controller to adaptively adjust the regularization coefficient. Furthermore, we provide a detailed algorithmic complexity analysis, showing that our method preserves the computational efficiency of stochastic gradient descent while improving adaptivity. Experimental results on real-world power load datasets validate the superiority of our method in both imputation accuracy and training efficiency compared to existing baselines.

cs.LG↗
arXiv2024

XL3M: A Training-free Framework for LLM Length Extension Based on Segment-wise Inference

Length generalization failure problem, namely the large language model (LLM) fails to generalize to texts longer than its maximum training length, greatly restricts the application of LLM in the scenarios with streaming long inputs. To address this problem, the existing methods either require substantial costs or introduce precision loss. In this paper, we empirically find that the accuracy of the LLM's prediction is highly correlated to its certainty. Based on this, we propose an efficient training free framework, named XL3M (it means extra-long large language model), which enables the LLMs trained on short sequences to reason extremely long sequence without any further training or fine-tuning. Under the XL3M framework, the input context will be firstly decomposed into multiple short sub-contexts, where each sub-context contains an independent segment and a common ``question'' which is a few tokens from the end of the original context. Then XL3M gives a method to measure the relevance between each segment and the ``question'', and constructs a concise key context by splicing all the relevant segments in chronological order. The key context is further used instead of the original context to complete the inference task. Evaluations on comprehensive benchmarks show the superiority of XL3M. Using our framework, a Llama2-7B model is able to reason 20M long sequences on an 8-card Huawei Ascend 910B NPU machine with 64GB memory per card.

cs.CL↗
arXiv2023

One for Multiple: Physics-informed Synthetic Data Boosts Generalizable Deep Learning for Fast MRI Reconstruction

Magnetic resonance imaging (MRI) is a widely used radiological modality renowned for its radiation-free, comprehensive insights into the human body, facilitating medical diagnoses. However, the drawback of prolonged scan times hinders its accessibility. The k-space undersampling offers a solution, yet the resultant artifacts necessitate meticulous removal during image reconstruction. Although Deep Learning (DL) has proven effective for fast MRI image reconstruction, its broader applicability across various imaging scenarios has been constrained. Challenges include the high cost and privacy restrictions associated with acquiring large-scale, diverse training data, coupled with the inherent difficulty of addressing mismatches between training and target data in existing DL methodologies. Here, we present a novel Physics-Informed Synthetic data learning framework for Fast MRI, called PISF. PISF marks a breakthrough by enabling generalized DL for multi-scenario MRI reconstruction through a single trained model. Our approach separates the reconstruction of a 2D image into many 1D basic problems, commencing with 1D data synthesis to facilitate generalization. We demonstrate that training DL models on synthetic data, coupled with enhanced learning techniques, yields in vivo MRI reconstructions comparable to or surpassing those of models trained on matched realistic datasets, reducing the reliance on real-world MRI data by up to 96%. Additionally, PISF exhibits remarkable generalizability across multiple vendors and imaging centers. Its adaptability to diverse patient populations has been validated through evaluations by ten experienced medical professionals. PISF presents a feasible and cost-effective way to significantly boost the widespread adoption of DL in various fast MRI applications.

eess.IV↗
arXiv2022

Zeroth-Order Stochastic Coordinate Methods for Decentralized Non-convex Optimization

In this letter, we first propose a \underline{Z}eroth-\underline{O}rder c\underline{O}ordinate \underline{M}ethod~(ZOOM) to solve the stochastic optimization problem over a decentralized network with only zeroth-order~(ZO) oracle feedback available. Moreover, we equip a simple mechanism "powerball" to ZOOM and propose ZOOM-PB to accelerate the convergence of ZOOM. Compared with the existing methods, we verify the proposed algorithms through two benchmark examples in the literature, namely the black-box binary classification and the generating adversarial examples from black-box DNNs in order to compare with the existing state-of-the-art centralized and distributed ZO algorithms. The numerical results demonstrate a faster convergence rate of the proposed algorithms.

math.OC↗
arXiv2021

One-dimensional Deep Low-rank and Sparse Network for Accelerated MRI

Deep learning has shown astonishing performance in accelerated magnetic resonance imaging (MRI). Most state-of-the-art deep learning reconstructions adopt the powerful convolutional neural network and perform 2D convolution since many magnetic resonance images or their corresponding k-space are in 2D. In this work, we present a new approach that explores the 1D convolution, making the deep network much easier to be trained and generalized. We further integrate the 1D convolution into the proposed deep network, named as One-dimensional Deep Low-rank and Sparse network (ODLS), which unrolls the iteration procedure of a low-rank and sparse reconstruction model. Extensive results on in vivo knee and brain datasets demonstrate that, the proposed ODLS is very suitable for the case of limited training subjects and provides improved reconstruction performance than state-of-the-art methods both visually and quantitatively. Additionally, ODLS also shows nice robustness to different undersampling scenarios and some mismatches between the training and test data. In summary, our work demonstrates that the 1D deep learning scheme is memory-efficient and robust in fast MRI.

eess.IV↗
arXiv2020

Optimal Rates of Distributed Regression with Imperfect Kernels

Distributed machine learning systems have been receiving increasing attentions for their efficiency to process large scale data. Many distributed frameworks have been proposed for different machine learning tasks. In this paper, we study the distributed kernel regression via the divide and conquer approach. This approach has been proved asymptotically minimax optimal if the kernel is perfectly selected so that the true regression function lies in the associated reproducing kernel Hilbert space. However, this is usually, if not always, impractical because kernels that can only be selected via prior knowledge or a tuning process are hardly perfect. Instead it is more common that the kernel is good enough but imperfect in the sense that the true regression can be well approximated by but does not lie exactly in the kernel space. We show distributed kernel regression can still achieves capacity independent optimal rate in this case. To this end, we first establish a general framework that allows to analyze distributed regression with response weighted base algorithms by bounding the error of such algorithms on a single data set, provided that the error bounds has factored the impact of the unexplained variance of the response variable. Then we perform a leave one out analysis of the kernel ridge regression and bias corrected kernel ridge regression, which in combination with the aforementioned framework allows us to derive sharp error bounds and capacity independent optimal rates for the associated distributed kernel regression algorithms. As a byproduct of the thorough analysis, we also prove the kernel ridge regression can achieve rates faster than $N^{-1}$ (where $N$ is the sample size) in the noise free setting which, to our best knowledge, are first observed and novel in regression learning.

cs.LG↗
arXiv2020

Quasi-convex subsets in Alexandrov spaces with lower curvature bound

In this paper, we introduce quasi-convex subsets in Alxandrov spaces with lower curvature bound, which include not only all closed convex subsets without boundary but also all extremal subsets. Moreover, we explore several essential properties of such kind of subsets including a generalized Liberman theorem. It turns out that the quasi-convex subset is a nice and fundamental concept to illustrate the similarities and differences between Riemannian manifolds and Alxandrov spaces with lower curvature bound.

math.MG↗
arXiv2015

An Isometrical ${\Bbb C\Bbb P}^{n}$-Theorem

Let $M^n\ (n\geq3)$ be a complete Riemannian manifold with $\sec_M\geq 1$, and let $M_i^{n_i}$ ($i=1,2$) be two comlplete totally geodesic submanifolds in $M$. We prove that if $n_1+n_2=n-2$ and if the distance $|M_1M_2|\geq\frac{\pi}{2}$, then $M_i$ is isometric to $\Bbb S^{n_i}/\Bbb Z_h$, ${\Bbb C\Bbb P}^{\frac {n_i}2}$ or ${\Bbb C\Bbb P}^{\frac {n_i}2}/\Bbb Z_2$ with the canonical metric when $n_i>0$, and thus $M$ is isometric to $\Bbb S^n/\Bbb Z_h$, ${\Bbb C\Bbb P}^{\frac n2}$ or ${\Bbb C\Bbb P}^{\frac n2}/\Bbb Z_2$ except possibly when $n=3$ and $M_1$ (or $M_2$) $\stackrel{\rm iso}{\cong}\Bbb S^{1}/\Bbb Z_h$ with $h\geq 2$ or $n=4$ and $M_1$ (or $M_2$) $\stackrel{\rm iso}{\cong}\Bbb{RP}^2$.

math.DG↗
arXiv2015

On the Blaschke's Conjecture

The Blaschke's conjecture asserts that if $\diam(M)=\text{Inj}(M)=\frac\pi2$ (up to a rescaling) for a complete Riemannian manifold $M$, then $M$ is isometric to $\Bbb S^n(\frac12)$, ${\Bbb R\Bbb P}^{n}$, ${\Bbb C\Bbb P}^{n}$, ${\Bbb H\Bbb P}^{n}$ or ${\Bbb Ca\Bbb P}^{2}$ endowed with the canonical metric. In the paper, we prove that the conjecture is true if we in addition assume that $\sec_M\geq1$.

math.DG↗
arXiv2014

On $\frac\pi2$-separated subsets of Alexandrov spaces with curvature $\geq1$

Let $M$ be an $n$-dimensional Alexandrov space with curvature $\geq 1$, and let $\{q_1,\cdots,q_k\}$ be any $\frac\pi2$-separated subset in $M$ (i.e. the distance $|q_iq_j|\geq\frac{\pi}{2}$ for any $i\neq j$). Under the additional conditions "$|q_iq_j|<\pi$" and "the diameter $\diam(M)\leq \frac\pi2$", we respectively give the upper bound of $k$ (which depends only on $n$), and we classify the (topological or geometric) structure of $M$ when $k$ attains the upper bound.

math.DG↗