arXiv subjects

Chi Ma

Publications and source records attributed to Chi Ma.

17 recordsLinked to original sources

arXiv2026

MTServe: Efficient Serving for Generative Recommendation Models with Hierarchical Caches

Generative recommendation (GR) offers superior modeling capabilities but suffers from prohibitive inference costs due to the repeated encoding of long user histories. While cross-request Key-Value (KV) cache reuse presents a significant optimization opportunity, the massive scale of individual user states creates a storage explosion that far exceeds physical GPU limits. We propose MTServe, a hierarchical cache management system that virtualizes GPU memory by leveraging host RAM as a scalable backup store. To bridge the I/O gap between tiers, MTServe introduces a suite of system-level optimizations, including a hybrid storage layout, an asynchronous data transfer pipeline, and a locality-driven replacement policy. On both public and production datasets, MTServe delivers up to 3.1* speedup while maintaining near-perfect hit ratios (>98.5%).

cs.LG↗

arXiv2026

Short Chains, Deep Thoughts: Balancing Reasoning Efficiency and Intra-Segment Capability via Split-Merge Optimization

While Large Reasoning Models (LRMs) have demonstrated impressive capabilities in solving complex tasks through the generation of long reasoning chains, this reliance on verbose generation results in significant latency and computational overhead. To address these challenges, we propose \textbf{CoSMo} (\textbf{Co}nsistency-Guided \textbf{S}plit-\textbf{M}erge \textbf{O}ptimization), a framework designed to eliminate structural redundancy rather than indiscriminately restricting token volume. Specifically, CoSMo utilizes a split-merge algorithm that dynamically refines reasoning chains by merging redundant segments and splitting logical gaps to ensure coherence. We then employ structure-aligned reinforcement learning with a novel segment-level budget to supervise the model in maintaining efficient reasoning structures throughout training. Extensive experiments across multiple benchmarks and backbones demonstrate that CoSMo achieves superior performance, improving accuracy by \textbf{3.3} points while reducing segment usage by \textbf{28.7\%} on average compared to reasoning efficiency baselines.

cs.CL↗

arXiv2025

MTGR: Industrial-Scale Generative Recommendation Framework in Meituan

Scaling law has been extensively validated in many domains such as natural language processing and computer vision. In the recommendation system, recent work has adopted generative recommendations to achieve scalability, but their generative approaches require abandoning the carefully constructed cross features of traditional recommendation models. We found that this approach significantly degrades model performance, and scaling up cannot compensate for it at all. In this paper, we propose MTGR (Meituan Generative Recommendation) to address this issue. MTGR is modeling based on the HSTU architecture and can retain the original deep learning recommendation model (DLRM) features, including cross features. Additionally, MTGR achieves training and inference acceleration through user-level compression to ensure efficient scaling. We also propose Group-Layer Normalization (GLN) to enhance the performance of encoding within different semantic spaces and the dynamic masking strategy to avoid information leakage. We further optimize the training frameworks, enabling support for our models with 10 to 100 times computational complexity compared to the DLRM, without significant cost increases. MTGR achieved 65x FLOPs for single-sample forward inference compared to the DLRM model, resulting in the largest gain in nearly two years both offline and online. This breakthrough was successfully deployed on Meituan, the world's largest food delivery platform, where it has been handling the main traffic.

cs.IR↗

arXiv2025

MTGenRec: An Efficient Distributed Training System for Generative Recommendation Models in Meituan

Recommendation is crucial for both user experience and company revenue in Meituan as a leading lifestyle company, and generative recommendation models (GRMs) are shown to produce quality recommendations recently. However, existing systems are limited by insufficient functionality support and inefficient implementations for training GRMs in industrial scenarios. As such, we introduce MTGenRec as an efficient and scalable system for GRM training. Specifically, to handle real-time insertions/deletions of sparse embeddings, MTGenRec employs dynamic hash tables to replace static ones. To improve training efficiency, MTGenRec conducts dynamic sequence balancing to address the computation load imbalances among GPUs and adopts feature ID deduplication alongside automatic table merging to accelerate embedding lookup. Extensive experiments show that MTGenRec improves training throughput by $1.6\times -- 2.4\times$ while achieving good scalability when running over 100 GPUs. MTGenRec has been deployed for many applications in Meituan and is now handling hundreds of millions of requests on a daily basis. On the delivery platform, we observe a 1.22% growth in user order volume and a 1.31% enhancement in online PV_CTR.

cs.DC↗

arXiv2025

HyperTree Planning: Enhancing LLM Reasoning via Hierarchical Thinking

Recent advancements have significantly enhanced the performance of large language models (LLMs) in tackling complex reasoning tasks, achieving notable success in domains like mathematical and logical reasoning. However, these methods encounter challenges with complex planning tasks, primarily due to extended reasoning steps, diverse constraints, and the challenge of handling multiple distinct sub-tasks. To address these challenges, we propose HyperTree Planning (HTP), a novel reasoning paradigm that constructs hypertree-structured planning outlines for effective planning. The hypertree structure enables LLMs to engage in hierarchical thinking by flexibly employing the divide-and-conquer strategy, effectively breaking down intricate reasoning steps, accommodating diverse constraints, and managing multiple distinct sub-tasks in a well-organized manner. We further introduce an autonomous planning framework that completes the planning process by iteratively refining and expanding the hypertree-structured planning outlines. Experiments demonstrate the effectiveness of HTP, achieving state-of-the-art accuracy on the TravelPlanner benchmark with Gemini-1.5-Pro, resulting in a 3.6 times performance improvement over o1-preview.

cs.AI↗

arXiv2025

Sparse Brains are Also Adaptive Brains: Cognitive-Load-Aware Dynamic Activation for LLMs

Dense large language models(LLMs) face critical efficiency bottlenecks as they rigidly activate all parameters regardless of input complexity. While existing sparsity methods(static pruning or dynamic activation) address this partially, they either lack adaptivity to contextual or model structural demands or incur prohibitive computational overhead. Inspired by human brain's dual-process mechanisms - predictive coding (N400) for backbone sparsity and structural reanalysis (P600) for complex context - we propose CLADA, a \textit{\textbf{C}ognitive-\textbf{L}oad-\textbf{A}ware \textbf{D}ynamic \textbf{A}ctivation} framework that synergizes statistical sparsity with semantic adaptability. Our key insight is that LLM activations exhibit two complementary patterns: 1) \textit{Global statistical sparsity} driven by sequence-level prefix information, and 2) \textit{Local semantic adaptability} modulated by cognitive load metrics(e.g., surprisal and entropy). CLADA employs a hierarchical thresholding strategy: a baseline from offline error-controlled optimization ensures 40\%+ sparsity, dynamically adjusted by real-time cognitive signals. Evaluations across six mainstream LLMs and nine benchmarks demonstrate that CLADA achieves \textbf{~20\% average speedup with <2\% accuracy drop}, outperforming Griffin (5\%+ degradation) and TT (negligible speedup). Crucially, we establish the first formal connection between neurolinguistic event-related potential (ERP) components and LLM efficiency mechanisms through multi-level regression analysis ($R^2=0.17$ for sparsity-adaptation synergy). Requiring no retraining or architectural changes, CLADA offers a deployable solution for resource-aware LLM inference while advancing biologically-inspired AI design. Our code is available at \href{https://github.com/Oldify/CLADA}{CLADA}.

cs.CL↗

arXiv2024

First Activations Matter: Training-Free Methods for Dynamic Activation in Large Language Models

Dynamic activation (DA) techniques, such as DejaVu and MoEfication, have demonstrated their potential to significantly enhance the inference efficiency of large language models (LLMs). However, these techniques often rely on ReLU activation functions or require additional parameters and training to maintain performance. This paper introduces a training-free Threshold-based Dynamic Activation(TDA) method that leverage sequence information to exploit the inherent sparsity of models across various architectures. This method is designed to accelerate generation speed by 18-25\% without significantly compromising task performance, thereby addressing the limitations of existing DA techniques. Moreover, we delve into the root causes of LLM sparsity and theoretically analyze two of its critical features: history-related activation uncertainty and semantic-irrelevant activation inertia. Our comprehensive analyses not only provide a robust theoretical foundation for DA methods but also offer valuable insights to guide future research in optimizing LLMs for greater efficiency and effectiveness.

cs.CL↗

arXiv2024

The differences in the origination and properties of the near-Earth solar wind between solar cycles 23 and 24

The dependence of the sources and properties of the near-Earth solar wind on solar cycle activity is an important issue in solar and space physics. We use the improved two-step mapping procedure that takes into account the initial acceleration processes to trace the near-Earth solar winds back to their source regions from 1999 to 2020, covering solar cycles (SCs) 23 and 24. Then the solar wind is categorized into coronal hole (CH), active region (AR), and quiet Sun (QS) solar wind based on the source region types. We find that the proportions of CH and AR (QS) wind during SC 23 are higher (lower) than those during SC 24. During solar maximum and declining phases, the magnetic field strength, speed, helium abundance (AHe), and charge states of all three types of solar wind during SC 23 are generally higher than those during SC 24. During solar minimum, these parameters of solar wind are generally lower during SC 23 than those during SC 24. There is a significant decrease in the charge states of all three types of solar wind during the solar minimum of SC 23. The present statistical results demonstrate that the sources and properties of the solar wind are both influenced by solar cycle amplitude. The temperatures of AR, QS, and CH regions exhibit significant difference at low altitudes, whereas they are almost uniform at high altitudes.

astro-ph.SR↗

arXiv2024

MOYU: A Theoretical Study on Massive Over-activation Yielded Uplifts in LLMs

Massive Over-activation Yielded Uplifts(MOYU) is an inherent property of large language models, and dynamic activation(DA) based on the MOYU property is a clever yet under-explored strategy designed to accelerate inference in these models. Existing methods that utilize MOYU often face a significant 'Impossible Trinity': struggling to simultaneously maintain model performance, enhance inference speed, and extend applicability across various architectures. Due to the theoretical ambiguities surrounding MOYU, this paper elucidates the root cause of the MOYU property and outlines the mechanisms behind two primary limitations encountered by current DA methods: 1) history-related activation uncertainty, and 2) semantic-irrelevant activation inertia. Our analysis not only underscores the limitations of current dynamic activation strategies within large-scale LLaMA models but also proposes opportunities for refining the design of future sparsity schemes.

cs.LG↗

arXiv2024

Dynamic Activation Pitfalls in LLaMA Models: An Empirical Study

In this work, we systematically investigate the efficacy of dynamic activation mechanisms within the LLaMA family of language models. Despite the potential of dynamic activation methods to reduce computation and increase speed in models using the ReLU activation function, our empirical findings have uncovered several inherent pitfalls in the current dynamic activation schemes. Through extensive experiments across various dynamic activation strategies, we demonstrate that LLaMA models usually underperform when compared to their ReLU counterparts, particularly in scenarios demanding high sparsity ratio. We attribute these deficiencies to a combination of factors: 1) the inherent complexity of dynamically predicting activation heads and neurons; 2) the inadequate sparsity resulting from activation functions; 3) the insufficient preservation of information resulting from KV cache skipping. Our analysis not only sheds light on the limitations of dynamic activation in the context of large-scale LLaMA models but also proposes roadmaps for enhancing the design of future sparsity schemes.

cs.LG↗

arXiv2024

Re-evaluating the Memory-balanced Pipeline Parallelism: BPipe

Pipeline parallelism is an essential technique in the training of large-scale Transformer models. However, it suffers from imbalanced memory consumption, leading to insufficient memory utilization. The BPipe technique was proposed to address this issue and has proven effective in the GPT-3 model. Nevertheless, our experiments have not yielded similar benefits for LLaMA training. Additionally, BPipe only yields negligible benefits for GPT-3 training when applying flash attention. We analyze the underlying causes of the divergent performance of BPipe on GPT-3 and LLaMA. Furthermore, we introduce a novel method to estimate the performance of BPipe.

cs.LG↗

arXiv2021

Metal-Insulator transition in strained Graphene: A quantum Monte carlo study

Motivated by the possibility of a strain tuning effect on electronic properties of graphene, the semimetal-Mott insulator transition process on the uniaxial honeycomb lattice is numerically studied using Determinant Quantum Monte Carlo. As our simulations are based on the half-filled repulsive Hubbard model, the system is sign problem free. Herein, the temperature-dependent DC conductivity is used to characterize electronic transport properties. The data suggest that metallic is suppressed in the presence of strain. More interestingly, within the finite-size scaling study, a novel antiferromagnetic phase arises at around $U\sim U_{c}$. Therefore, a phase diagram generated by the competition between interactions and strain is established, which may help to expand the application of strain effect on graphene.

cond-mat.str-el↗

arXiv2021

Morphological Operation Residual Blocks: Enhancing 3D Morphological Feature Representation in Convolutional Neural Networks for Semantic Segmentation of Medical Images

The shapes and morphology of the organs and tissues are important prior knowledge in medical imaging recognition and segmentation. The morphological operation is a well-known method for morphological feature extraction. As the morphological operation is performed well in hand-crafted image segmentation techniques, it is also promising to design an approach to approximate morphological operation in the convolutional networks. However, using the traditional convolutional neural network as a black-box is usually hard to specify the morphological operation action. Here, we introduced a 3D morphological operation residual block to extract morphological features in end-to-end deep learning models for semantic segmentation. This study proposed a novel network block architecture that embedded the morphological operation as an infinitely strong prior in the convolutional neural network. Several 3D deep learning models with the proposed morphological operation block were built and compared in different medical imaging segmentation tasks. Experimental results showed the proposed network achieved a relatively higher performance in the segmentation tasks comparing with the conventional approach. In conclusion, the novel network block could be easily embedded in traditional networks and efficiently reinforce the deep learning models for medical imaging segmentation.

cs.CV↗

arXiv2020

Characteristics of Pumping Current in a YBCO Coil by a Pulse-Type Magnetic Flux Pump

2G high temperature superconducting (HTS) wires, YBCO coated conductors, perform a better carrying current capability, which is potentially applied in the manufacture of HTS magnets. This paper presents the experimental results of the pumping current for YBCO coils using a pulse-type magnetic flux pump in the conduction-cooling system and liquid nitrogen bath (LN2) cryogenic environment. Optimization of the flux pump used in the conduction-cooling system is that a constantan heater was added to keep the temperature of the pumping bridge at a certain value. Excitation effects of the YBCO coil at different temperatures were investigated in the conduction-cooling system. A fast-increasing of pumping current in the YBCO coil occurs when the temperature of the YBCO sheet (i.e., pumping bridge) is in the range of 50 K to 80 K. The relationships between saturated pumping current and input voltage, working frequency, numbers of magnetic poles were also studied. Using the seven-pole configuration, the saturated current can reach 155 A when the frequency is 20 Hz and the voltage is 6 V. The excitation characteristics of the flux pump in the LN2 cooling system show the possibility of the pulse-type magnetic flux pump for the practical application of HTS magnets.

physics.app-ph↗

arXiv2019

Quantum confinement and edge effects on electronic properties of zigzag green phosphorene nanoribbons

First principles density-functional theory calculations were performed to investigate quantum confinement and edge effects on the electronic properties of zigzag green phosphorene nanoribbons (ZGPNRs) with edge chemical species including H, OH, F, Cl, O, and S for the ribbons width in the range of 0.5 \~{} 3.7 nm. The ZGPNRs were obtained from the relaxed two-dimensional (2D) green phosphorene monolayer with different cutting strategies and the most energetically favorable ribbon configuration was selected for further exploration of the size and edge effects. It was found that the electronic properties of the ZGPNRs are strongly associated with the ribbon width and edge chemical species. They show either semiconducting or metallic features depending on the edge functionalization species. The ZGPNRs show semiconducting behavior with the edge species of H, OH, F, or Cl (Group \uppercase\expandafter{\romannumeral1}), while exhibit metallic characteristics with pristine or O, S edges (Group \uppercase\expandafter{\romannumeral2}). The conduction band minimum (CBM) and valence band maximum (VBM) of the ZGPNRs with the Group \uppercase\expandafter{\romannumeral1} edge are primarily located at the inner P atoms and the edge P and functionalization atoms have little contribution. However, for the Group \uppercase\expandafter{\romannumeral2} edge, the electronic bands crossing the Fermi level are dominantly contributed by the edge atoms. It was also found that the band gap and work function of the ZGPNRs are tunable by varying ribbon width and edge functionalization species.

cond-mat.str-el↗

arXiv2019

AMAD: Adversarial Multiscale Anomaly Detection on High-Dimensional and Time-Evolving Categorical Data

Anomaly detection is facing with emerging challenges in many important industry domains, such as cyber security and online recommendation and advertising. The recent trend in these areas calls for anomaly detection on time-evolving data with high-dimensional categorical features without labeled samples. Also, there is an increasing demand for identifying and monitoring irregular patterns at multiple resolutions. In this work, we propose a unified end-to-end approach to solve these challenges by combining the advantages of Adversarial Autoencoder and Recurrent Neural Network. The model learns data representations cross different scales with attention mechanisms, on which an enhanced two-resolution anomaly detector is developed for both instances and data blocks. Extensive experiments are performed over three types of datasets to demonstrate the efficacy of our method and its superiority over the state-of-art approaches.

cs.LG↗

arXiv2010

Neutron-rich chromium isotope anomalies in supernova nanoparticles

Neutron-rich isotopes with masses near that of iron are produced in type Ia and II supernovae. Traces of such nucleosynthesis are found in primitive meteorites in the form of variations in the isotopic abundance of 54Cr, the most neutron-rich stable isotope of chromium. The hosts of these isotopic anomalies must be presolar grains that condensed in the outflows of supernovae, offering the opportunity to study the nucleosynthesis of iron-peak nuclei in ways that complement spectroscopic observations and can inform models of stellar evolution. However, despite almost two decades of extensive search, the carrier of 54Cr anomalies is still unknown, presumably because it is fine-grained and is chemically labile. Here we identify in the primitive meteorite Orgueil the carrier of 54Cr-anomalies as nanoparticles, most likely spinels that show large enrichments in 54Cr relative to solar composition (54Cr/52Cr ratio >3.6xsolar). Such large enrichments in 54Cr can only be produced in supernovae. The mineralogy of the grains supports condensation in the O/Ne-O/C zones of a type II supernova, although a type Ia origin cannot be excluded. We suggest that planetary materials incorporated different amounts of these nanoparticles, possibly due to late injection by a nearby supernova that also delivered 26Al and 60Fe to the solar system. This idea explains why the relative abundance of 54Cr and other neutron-rich isotopes vary between planets and meteorites. We anticipate that future isotopic studies of the grains identified here will shed new light on the birth of the solar system and the conditions insupernovae.

astro-ph.SR↗