arXiv subjects

Shengnan Wang

Publications and source records attributed to Shengnan Wang.

16 recordsLinked to original sources

arXiv2026

Accelerating Long-Tail Generation in Synchronous RLHF Training via Adaptive Tensor Parallelism

Reinforcement Learning from Human Feedback (RLHF) has become a key post-training paradigm for improving model quality. However, the synchronous three-stage RLHF pipeline is often bottlenecked by the generation stage, where response-length skew causes the effective batch size to shrink rapidly during decoding, leaving GPUs underutilized while a few long responses remain unfinished. Mainstream frameworks employ a static tensor parallelism (TP) configuration that cannot adapt to changing batch characteristics, leaving substantial performance headroom unexplored. We propose PAT, an adaptive TP method that dynamically reconfigures TP during the generation stage of each RLHF iteration. PAT introduces two key techniques. First, a predictor-guided online reconfiguration method decides both the reconfiguration point and the target TP configuration based on offline profiling, triggering reconfiguration only when the predicted latency benefit outweighs the reconfiguration overhead. Second, a lightweight online reconfiguration mechanism updates only the states and layouts affected by TP changes: it adapts unfinished decoding states through a cost-model-based choice between KV-cache migration and recomputation, performs in-place weight resharding, and reuses cached communication groups. We implement PAT on top of SGLang and integrate it with the VeRL framework. Evaluations on LLaMA3.1-8B and Qwen3-14B using DeepScaleR show that PAT reduces generation latency by up to 34.6% and end-to-end RLHF training iteration latency by up to 27.2% compared to the original VeRL setup.

cs.AI↗

arXiv2026

AdaCluster: Adaptive Query-Key Clustering for Sparse Attention in Video Generation

Video diffusion transformers (DiTs) suffer from prohibitive inference latency due to quadratic attention complexity. Existing sparse attention methods either overlook semantic similarity or fail to adapt to heterogeneous token distributions across layers, leading to model performance degradation. We propose AdaCluster, a training-free adaptive clustering framework that accelerates the generation of DiTs while preserving accuracy. AdaCluster applies an angle-similarity-preserving clustering method to query vectors for higher compression, and designs a euclidean-similarity-preserving clustering method for keys, covering cluster number assignment, threshold-wise adaptive clustering, and efficient critical cluster selection. Experiments on CogVideoX-2B, HunyuanVideo, and Wan-2.1 on one A40 GPU demonstrate up to 1.67-4.31x speedup with negligible quality degradation.

cs.CV↗

arXiv2026

LiteCache: A Query Similarity-Driven, GPU-Centric KVCache Subsystem for Efficient LLM Inference

During LLM inference, KVCache memory usage grows linearly with sequence length and batch size and often exceeds GPU capacity. Recent proposals offload KV states to host memory and reduce transfers using top-k attention. But their CPU-centric management of the on-GPU cache and CPU-GPU data movement incurs high overhead and fragments the bulk GPU execution that CUDA Graph relies on. To close this gap, we observe that adjacent queries within the same attention head exhibit strong directional similarity and retrieve highly overlapping top-k KV states. This insight enables a simple head granularity cache algorithm, QSAC, in which each head reuses its previously cached KV states whenever the current query is sufficiently similar to the prior one. QSAC further simplifies cache management primitives and cuts CPU involvement almost entirely. We develop LiteCache, a KVCache subsystem that incorporates QSAC. LiteCache introduces a GPU-centric synchronization controller and speculative sparse prefetching, enabling fully overlapped data movement and computation. These mechanisms produce a stable and predictable execution pattern that remains compatible with the bulk execution mode required by CUDA Graphs. Evaluation on two widely-used LLMs indicates that LiteCache achieves comparable accuracy to baselines, while sharply minimizing CPU overhead, fully utilizing PCIe bandwidth, thus improving decoding throughput by 10.7-224.2% on both H100 and A40 GPUs and easily supporting sequence lengths beyond 1M. We opensource LiteCache at https://anonymous.4open.science/r/LiteCache-888D.

cs.LG↗

arXiv2026

IceBench-S2S: A Benchmark of Deep Learning for Challenging Subseasonal-to-Seasonal Daily Arctic Sea Ice Forecasting in Deep Latent Space

Arctic sea ice plays a critical role in regulating Earth's climate system, significantly influencing polar ecological stability and human activities in coastal regions. Recent advances in artificial intelligence have facilitated the development of skillful pan-Arctic sea ice forecasting systems, where data-driven approaches showcase tremendous potential to outperform conventional physics-based numerical models in terms of accuracy, computational efficiency and forecasting lead times. Despite the latest progress made by deep learning (DL) forecasting models, most of their skillful forecasting lead times are confined to daily subseasonal scale and monthly averaged values for up to six months, which drastically hinders their deployment for real-world applications, e.g., maritime routine planning for Arctic transportation and scientific investigation. Extending daily forecasts from subseasonal to seasonal (S2S) scale is scientifically crucial for operational applications. To bridge the gap between the forecasting lead time of current DL models and the significant daily S2S scale, we introduce IceBench-S2S, the first comprehensive benchmark for evaluating DL approaches in mitigating the challenge of forecasting Arctic sea ice concentration in successive 180-day periods. It proposes a generalized framework that first compresses spatial features of daily sea ice data into a deep latent space. The temporally concatenated deep features are subsequently modeled by DL-based forecasting backbones to predict the sea ice variation at S2S scale. IceBench-S2S provides a unified training and evaluation pipeline for different backbones, along with practical guidance for model selection in polar environmental monitoring tasks.

cs.LG↗

arXiv2025

CentroidKV: Efficient Long-Context LLM Inference via KV Cache Clustering

Large language models (LLMs) with extended context windows have become increasingly prevalent for tackling complex tasks. However, the substantial Key-Value (KV) cache required for long-context LLMs poses significant deployment challenges. Existing approaches either discard potentially critical information needed for future generations or offer limited efficiency gains due to high computational overhead. In this paper, we introduce CentroidKV, a simple yet effective framework for online KV cache clustering. Our approach is based on the observation that key states exhibit high similarity along the sequence dimension. To enable efficient clustering, we divide the sequence into chunks and propose Chunked Soft Matching, which employs an alternating partition strategy within each chunk and identifies clusters based on similarity. CentroidKV then merges the KV cache within each cluster into a single centroid. Additionally, we provide a theoretical analysis of the computational complexity and the optimality of the intra-chunk partitioning strategy. Extensive experiments across various models and long-context benchmarks demonstrate that CentroidKV achieves up to 75% reduction in KV cache memory usage while maintaining comparable model performance. Moreover, with minimal computational overhead, CentroidKV accelerates the decoding stage of inference by up to $1.92\times$ and increases the serving throughput by up to $4\times$.

cs.CL↗

arXiv2025

HATA: Trainable and Hardware-Efficient Hash-Aware Top-k Attention for Scalable Large Model Inference

Large Language Models (LLMs) have emerged as a pivotal research area, yet the attention module remains a critical bottleneck in LLM inference, even with techniques like KVCache to mitigate redundant computations. While various top-$k$ attention mechanisms have been proposed to accelerate LLM inference by exploiting the inherent sparsity of attention, they often struggled to strike a balance between efficiency and accuracy. In this paper, we introduce HATA (Hash-Aware Top-$k$ Attention), a novel approach that systematically integrates low-overhead learning-to-hash techniques into the Top-$k$ attention process. Different from the existing top-k attention methods which are devoted to seeking an absolute estimation of qk score, typically with a great cost, HATA maps queries and keys into binary hash codes, and acquires the relative qk score order with a quite low cost, which is sufficient for realizing top-k attention. Extensive experiments demonstrate that HATA achieves up to 7.2$\times$ speedup compared to vanilla full attention while maintaining model accuracy. In addition, HATA outperforms the state-of-the-art top-$k$ attention methods in both accuracy and efficiency across multiple mainstream LLM models and diverse tasks. HATA is open source at https://github.com/gpzlx1/HATA.

cs.LG↗

arXiv2025

BigMac: A Communication-Efficient Mixture-of-Experts Model Structure for Fast Training and Inference

The Mixture-of-Experts (MoE) structure scales the Transformer-based large language models (LLMs) and improves their performance with only the sub-linear increase in computation resources. Recently, a fine-grained DeepSeekMoE structure is proposed, which can further improve the computing efficiency of MoE without performance degradation. However, the All-to-All communication introduced by MoE has become a bottleneck, especially for the fine-grained structure, which typically involves and activates more experts, hence contributing to heavier communication overhead. In this paper, we propose a novel MoE structure named BigMac, which is also fine-grained but with high communication efficiency. The innovation of BigMac is mainly due to that we abandon the \textbf{c}ommunicate-\textbf{d}escend-\textbf{a}scend-\textbf{c}ommunicate (CDAC) manner used by fine-grained MoE, which leads to the All-to-All communication always taking place at the highest dimension. Instead, BigMac designs an efficient \textbf{d}escend-\textbf{c}ommunicate-\textbf{c}ommunicate-\textbf{a}scend (DCCA) manner. Specifically, we add a descending and ascending projection at the entrance and exit of the expert, respectively, which enables the communication to perform at a very low dimension. Furthermore, to adapt to DCCA, we re-design the structure of small experts, ensuring that the expert in BigMac has enough complexity to address tokens. Experimental results show that BigMac achieves comparable or even better model quality than fine-grained MoEs with the same number of experts and a similar number of total parameters. Equally importantly, BigMac reduces the end-to-end latency by up to 3.09$\times$ for training and increases the throughput by up to 3.11$\times$ for inference on state-of-the-art AI computing frameworks including Megatron, Tutel, and DeepSpeed-Inference.

cs.LG↗

arXiv2024

SIFM: A Foundation Model for Multi-granularity Arctic Sea Ice Forecasting

Arctic sea ice performs a vital role in global climate and has paramount impacts on both polar ecosystems and coastal communities. In the last few years, multiple deep learning based pan-Arctic sea ice concentration (SIC) forecasting methods have emerged and showcased superior performance over physics-based dynamical models. However, previous methods forecast SIC at a fixed temporal granularity, e.g. sub-seasonal or seasonal, thus only leveraging inter-granularity information and overlooking the plentiful inter-granularity correlations. SIC at various temporal granularities exhibits cumulative effects and are naturally consistent, with short-term fluctuations potentially impacting long-term trends and long-term trends provides effective hints for facilitating short-term forecasts in Arctic sea ice. Therefore, in this study, we propose to cultivate temporal multi-granularity that naturally derived from Arctic sea ice reanalysis data and provide a unified perspective for modeling SIC via our Sea Ice Foundation Model. SIFM is delicately designed to leverage both intra-granularity and inter-granularity information for capturing granularity-consistent representations that promote forecasting skills. Our extensive experiments show that SIFM outperforms off-the-shelf deep learning models for their specific temporal granularity.

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

Digital twin-assisted three-dimensional electrical capacitance tomography for multiphase flow imaging

Three-dimensional electrical capacitance tomography (3D-ECT) has shown promise for visualizing industrial multiphase flows. However, existing 3D-ECT approaches suffer from limited imaging resolution and lack assessment metrics, hampering their effectiveness in quantitative multiphase flow imaging. This paper presents a digital twin (DT)-assisted 3D-ECT, aiming to overcome these limitations and enhance our understanding of multiphase flow dynamics. The DT framework incorporates a 3D fluid-electrostatic field coupling model (3D-FECM) that digitally represents the physical 3D-ECT system, which enables us to simulate real multiphase flows and generate a comprehensive virtual multiphase flow 3D imaging dataset. Additionally, the framework includes a deep neural network named 3D deep back projection (3D-DBP), which learns multiphase flow features from the virtual dataset and enables more accurate 3D flow imaging in the real world. The DT-assisted 3D-ECT was validated through virtual and physical experiments, demonstrating superior image quality, noise robustness and computational efficiency compared to conventional 3D-ECT approaches. This research contributes to developing accurate and reliable 3D-ECT techniques and their implementation in multiphase flow systems across various industries.

eess.IV↗

arXiv2021

Digital Twin of Electrical Tomography for Quantitative Multiphase Flow Imaging

We report a digital twin (DT) framework of electrical tomography (ET) to address the challenge of real-time quantitative multiphase flow imaging based on non-invasive and non-radioactive technologies. Multiphase flow is ubiquitous in nature, industry, and research. Accurate flow imaging is the key to understanding this complex phenomenon. Existing non-radioactive multiphase flow imaging methods based on electrical tomography are limited to providing qualitative images. The proposed DT framework, building upon a synergistic integration of 3D field coupling simulation, model-based deep learning, and edge computing, allows ET to dynamically learn the flow features in the virtual space and implement the model in the physical system, thus providing unprecedented resolution and accuracy. The DT framework is demonstrated on gas-liquid two-phase flow and electrical capacitance tomography (ECT). It can be readily extended to various tomography modalities, scenarios, and scales in biomedical, energy, and aerospace applications as an effective alternative to radioactive solutions for precise flow visualization and characterization.

physics.flu-dyn↗

arXiv2021

CoRe: An Efficient Coarse-refined Training Framework for BERT

In recent years, BERT has made significant breakthroughs on many natural language processing tasks and attracted great attentions. Despite its accuracy gains, the BERT model generally involves a huge number of parameters and needs to be trained on massive datasets, so training such a model is computationally very challenging and time-consuming. Hence, training efficiency should be a critical issue. In this paper, we propose a novel coarse-refined training framework named CoRe to speed up the training of BERT. Specifically, we decompose the training process of BERT into two phases. In the first phase, by introducing fast attention mechanism and decomposing the large parameters in the feed-forward network sub-layer, we construct a relaxed BERT model which has much less parameters and much lower model complexity than the original BERT, so the relaxed model can be quickly trained. In the second phase, we transform the trained relaxed BERT model into the original BERT and further retrain the model. Thanks to the desired initialization provided by the relaxed model, the retraining phase requires much less training steps, compared with training an original BERT model from scratch with a random initialization. Experimental results show that the proposed CoRe framework can greatly reduce the training time without reducing the performance.

cs.CL↗

arXiv2020

Progressively Stacking 2.0: A Multi-stage Layerwise Training Method for BERT Training Speedup

Pre-trained language models, such as BERT, have achieved significant accuracy gain in many natural language processing tasks. Despite its effectiveness, the huge number of parameters makes training a BERT model computationally very challenging. In this paper, we propose an efficient multi-stage layerwise training (MSLT) approach to reduce the training time of BERT. We decompose the whole training process into several stages. The training is started from a small model with only a few encoder layers and we gradually increase the depth of the model by adding new encoder layers. At each stage, we only train the top (near the output layer) few encoder layers which are newly added. The parameters of the other layers which have been trained in the previous stages will not be updated in the current stage. In BERT training, the backward computation is much more time-consuming than the forward computation, especially in the distributed training setting in which the backward computation time further includes the communication time for gradient synchronization. In the proposed training strategy, only top few layers participate in backward computation, while most layers only participate in forward computation. Hence both the computation and communication efficiencies are greatly improved. Experimental results show that the proposed method can achieve more than 110% training speedup without significant performance degradation.

cs.CL↗

arXiv2019

High-Temperature Superconductivity in the Ti--H System at High Pressures

Search for stable high-pressure compounds in the Ti--H system reveals the existence of titanium hydrides with new stoichiometries, including Ibam-Ti$_2$H$_5$, I4/m-Ti$_5$H$_{13}$, I$\bar{4}$-Ti$_5$H$_{14}$, Fddd-TiH$_4$, Immm-Ti$_2$H$_{13}$, P$\bar{1}$-TiH$_{12}$, and C2/m-TiH$_{22}$. Our calculations predict I4/mmm $\rightarrow$ R$\bar{3}$m and I4/mmm $\rightarrow$ Cmma transitions in TiH and TiH$_2$, respectively. Phonons and the electron--phonon coupling of all searched titanium hydrides are analyzed at high pressure. It is found that Immm-Ti$_2$H$_{13}$ rather than the highest hydrogen content C2/m-TiH$_{22}$, exhibits the highest superconducting critical temperature T$_{c}$. The estimated T$_{c}$ of Immm-Ti$_2$H$_{13}$ and C2/m-TiH$_{22}$ are respectively 127.4--149.4 K ($μ^{*}$=0.1-0.15) at 350 GPa and 91.3--110.2 K at 250 GPa by numerically solving the Eliashberg equations. One of the effects of pressure on T$_{c}$ can be attributed to the softening and hardening of phonons with increasing pressure.

cond-mat.mtrl-sci↗

arXiv2015

Superconductivity of novel tin hydrides (Sn$_n$H$_m$) under pressure

With the motivation of discovering high-temperature superconductors, evolutionary algorithm is employed to search for all stable compounds in the Sn-H system. In addition to the traditional SnH$_4$, new hydrides SnH$_8$, SnH$_{12}$ and SnH$_{14}$ are found to be thermodynamically stable at high pressure. Dynamical stability and superconductivity of tin-hydrides are systematically investigated. I$\bar{4}$m2-SnH$_8$, C2/m-SnH$_{12}$ and C2/m-SnH$_{14}$ exhibit higher superconducting transition temperatures of 81, 93 and 97 K compared to the traditional compound SnH$_4$ with T$_c$ of 52 K at 200 GPa. An interesting bent H$_3^-$ in I$\bar{4}$m2-SnH$_8$ and novel liner H$_4^-$ in C2/m-SnH$_{12}$ are observed. All the new tin-hydrides remain metallic over their predicted range of stability. The intermediate-frequency wagging and bending vibrations have more contribution to electron-phonon coupling parameter than high-frequency stretching vibrations of H$_2$ and H$_3$.

cond-mat.supr-con↗

arXiv2015

Prediction of new thermodynamically stable aluminum oxides

Recently, it has been shown that under pressure, unexpected and counterintuitive chemical compounds become stable. Laser shock experiments (A. Rode, unpublished) on alumina (Al2O3) have shown non-equilibrium decomposition of alumina with the formation of free Al and a mysterious transparent phase. Inspired by these observations, with have explored the possibility of the formation of new chemical compounds in the system Al-O. Using the variable-composition structure prediction algorithm USPEX, in addition to the well-known Al2O3, we have found two extraordinary compounds Al4O7 and AlO2 to be thermodynamically stable in the pressure range 330-443 GPa and above 332 GPa, respectively. Both of these compounds at the same time contain oxide O2- and peroxide O22- ions, and both are insulating. Peroxo-groups are responsible for gap states, which significantly reduce the electronic band gap of both Al4O7 and AlO2.

physics.comp-ph↗