arXiv subjects

Rong Shi

Publications and source records attributed to Rong Shi.

12 recordsLinked to original sources

arXiv2026

PACE: Adaptive Budget Allocation for Time-Efficient Embodied Planning

Reasoning-enhanced large language models have achieved remarkable improvements in planning tasks, yet their deployment in embodied systems remains impractical due to prohibitive inference delays-often exceeding minutes per planning instance. The fundamental bottleneck stems from the serial nature of existing paradigms: models must complete all reasoning before any action execution, leaving execution time windows entirely unexploited. We introduce PACE (Planning with Adaptive Cognitive Effort), a framework that enables interleaved reasoning and execution through two key innovations: an Interleaved Think-Act architecture that pipelines cognitive processing with action execution, and a Dynamic Budget Allocator that adapts reasoning token budgets to available execution time windows. On the Robotouille benchmark using Qwen3-8B-AWQ, PACE achieves a 10% success rate-representing a 67% improvement over the ReAct+Think baseline-while delivering 6.9 times acceleration in thinking time compared to unconstrained reasoning. The framework hides 66.8% of thinking time within execution windows, demonstrating that strategic cognitive effort allocation can simultaneously improve both planning quality and time efficiency. These results provide evidence that time-aware architectural innovations enable reasoning models to operate in latency-sensitive embodied domains where they were previously impractical.

cs.RO↗

arXiv2026

MoWorld: A Flash World Model

The future of World Models depends not only on scaling model capability, but also on scaling practicality and inference efficiency. High-frame-rate inference enables responsive perception, planning, and control in real-world autonomous systems. To this end, we present MoWorld, a cost-effective yet high-performance Flash World Model with an end-to-end framework spanning data generation, pre-training, distillation, and efficient inference, enabling up to 50 FPS real-time interaction with cinematic visual quality without the need of high-end GPUs. To enable large-scale real-world deployment, MoWorld jointly optimizes model capability and cost throughout the entire development pipeline. Specifically, unlike existing approaches that primarily rely on large-scale video corpora, MoWorld is built upon a scalable 3D-native data engine accumulated from our large-scale 3D vision and generative modeling pipeline, enabling the efficient construction of geometrically consistent training data across diverse real-world and synthetic environments. Based on this foundation, a curriculum cross-frame pre-training strategy for stable and scalable World Model learning, an efficient denoising-step distillation algorithm to reduce diffusion training cost, and a mixed-precision parallel inference framework for low-cost real-time deployment. MoWorld is the first real-time interactive World Model built on the Neural Processing Unit (NPU) and can achieves up to 50 FPS in such the devices, enabling practical and efficient deployment at scale. Comprehensive evaluations demonstrate that MoWorld achieves leading performance; notably, its average inference cost is only 30\%-50\% of that of existing World Models, providing a practical foundation for large-scale real-world applications of World Models. We also demonstrate diverse applications of MoWorld.

cs.CV↗

arXiv2026

Low-scaling \textit{GW} calculations of quasi-particle energies for extended systems within the numerical atomic orbital framework

The many-body perturbation theory within the $GW$ approximation is a widely used method for describing the electronic band structures in real materials. Its application to large-scale systems is, however, impeded by its high computational cost. The rate-limiting steps in a typical $GW$ implementation are the evaluation of the polarization function under the random phase approximation (RPA) and the evaluation of the $GW$ self-energy, both of which have a canonical $O(N^4)$ scaling with $N$ being the system size. The conventional space-time algorithm within the plane-wave basis sets reduces the scaling from $O(N^4)$ to $O(N^3)$, albeit with a large prefactor and increased memory cost. Here, we present a space-time algorithm within the numerical atomic orbital (NAO) basis-set framework, for which the evaluation of the polarization function and self-energy is formally reduced to $O(N^2)$ or better with respect to system size. This is achieved by computing these quantities in real space, where low-scaling algorithms can be formulated by leveraging the localized resolution of identity (LRI) technique. The resulting NAO-based, LRI-enhanced space-time $GW$ algorithm has been implemented in the LibRPA library interfaced with the FHI-aims code package. Benchmark calculations for crystalline solids show that the low-scaling implementation yields quasi-particle energies in close agreement with the conventional $O(N^4)$ k-space formalism previously implemented in FHI-aims. For the systems studied here, the observed overall scaling is substantially reduced relative to the canonical approach, and the low-scaling implementation becomes advantageous already for systems containing fewer than 100 atoms.

cond-mat.mtrl-sci↗

arXiv2026

Seed2Scale: A Self-Evolving Data Engine for Embodied AI via Small to Large Model Synergy and Multimodal Evaluation

Existing data generation methods suffer from exploration limits, embodiment gaps, and low signal-to-noise ratios, leading to performance degradation during self-iteration. To address these challenges, we propose Seed2Scale, a self-evolving data engine that overcomes the data bottleneck through a heterogeneous synergy of "small-model collection, large-model evaluation, and target-model learning". Starting with as few as four seed demonstrations, the engine employs the lightweight Vision-Language-Action model, SuperTiny, as a dedicated collector, leveraging its strong inductive bias for robust exploration in parallel environments. Concurrently, a pre-trained Vision-Language Model is integrated as a Verifer to autonomously perform success/failure judgment and quality scoring for the massive generated trajectories. Seed2Scale effectively mitigates model collapse, ensuring the stability of the self-evolution process. Experimental results demonstrate that Seed2Scale exhibits signifcant scaling potential: as iterations progress, the success rate of the target model shows a robust upward trend, achieving a performance improvement of 131.2%. Furthermore, Seed2Scale signifcantly outperforms existing data augmentation methods, providing a scalable and cost-effective pathway for the large-scale development of Generalist Embodied AI. Project page: https://terminators2025.github.io/Seed2Scale.github.io

cs.RO↗

arXiv2025

MemFine: Memory-Aware Fine-Grained Scheduling for MoE Training

The training of large-scale Mixture of Experts (MoE) models faces a critical memory bottleneck due to severe load imbalance caused by dynamic token routing. This imbalance leads to memory overflow on GPUs with limited capacity, constraining model scalability. Existing load balancing methods, which cap expert capacity, compromise model accuracy and fail on memory-constrained hardware. To address this, we propose MemFine, a memory-aware fine-grained scheduling framework for MoE training. MemFine decomposes the token distribution and expert computation into manageable chunks and employs a chunked recomputation strategy, dynamically optimized through a theoretical memory model to balance memory efficiency and throughput. Experiments demonstrate that MemFine reduces activation memory by 48.03% and improves throughput by 4.42% compared to full recomputation-based baselines, enabling stable large-scale MoE training on memory-limited GPUs.

cs.DC↗

arXiv2025

MoFa: A Unified Performance Modeling Framework for LLM Pretraining

The exponential growth in LLM scales, with parameters soaring from billions to trillions, has necessitated distributed pretraining across large clusters comprising thousands to tens of thousands of devices. While hybrid parallelization strategies enable such pretraining, the vast combinatorial strategy space introduces significant optimization challenges. Traditional manual tuning methods incur prohibitive trial-and-error costs, and existing performance modeling approaches exhibit critical limitations: they fail to comprehensively account for prevalent optimization features and ignore the substantial overhead imposed by essential fault tolerance mechanisms like checkpoint recovery in long-duration pretraining. To address these gaps, we propose MoFa, a novel pretraining performance modeling framework that unifies multi-dimensional optimization features and fault tolerance. MoFa incorporates an enhanced cost model to accurately capture the effects of key optimizations and integrates a fault tolerance model based on historical cluster reliability data. Besides, a MoFa-based tuning system is developed to explore optimal pretraining performance and potential bottlenecks in various scenarios. Extensive modeling evaluations demonstrate that MoFa can achieve high prediction accuracy across various scenarios. In addition, through comprehensive tuning experiments, our framework systematically reveals the key factors influencing pretraining performance under different configurations, which provides solid a priori guidance for LLM pretraining system design and deployment.

cs.DC↗

arXiv2025

Disaggregated Prefill and Decoding Inference System for Large Language Model Serving on Multi-Vendor GPUs

LLM-based applications have been widely used in various industries, but with the increasing of models size, an efficient large language model (LLM) inference system is an urgent problem to be solved for service providers. Since the inference system is divided into two stage with different characteristics: Prefill and Decode, the two stage will interfere with each other during the inference process. Toward this end, a P-D disaggregated inference framework is proposed by some researchers. Current research is done on homogeneous GPUs, and lacks deployment solutions based on business scenarios. Compared with homogeneous GPUs, using heterogeneous GPUs to construct inference systems can better improve resource utilization and reduce costs. Even if GPUs from different vendors are used to build inference systems, on the basis of reducing costs, the resource utilization rate can be improved and the dependence on a single vendor can be reduced. Therefore, a P-D disaggreagetd inference system based on heterogeneous GPUs is designed, and the heterogeneous compatible transmission module in the system is designed to address heterogeneous GPU data compatibility issues. Then, a joint optimization algorithm of parallel strategy and instance number allocation is proposed to obtain the deployment solutions. Finally, the experimental results show that the P-D disaggregated inference system can well solve the hybrid inference problem of heterogeneous GPUs from different vendors, and the joint optimization algorithm can obtain the optimal deployment solution.

cs.DC↗

arXiv2025

RealMirror: A Comprehensive, Open-Source Vision-Language-Action Platform for Embodied AI

The emerging field of Vision-Language-Action (VLA) for humanoid robots faces several fundamental challenges, including the high cost of data acquisition, the lack of a standardized benchmark, and the significant gap between simulation and the real world. To overcome these obstacles, we propose RealMirror, a comprehensive, open-source embodied AI VLA platform. RealMirror builds an efficient, low-cost data collection, model training, and inference system that enables end-to-end VLA research without requiring a real robot. To facilitate model evolution and fair comparison, we also introduce a dedicated VLA benchmark for humanoid robots, featuring multiple scenarios, extensive trajectories, and various VLA models. Furthermore, by integrating generative models and 3D Gaussian Splatting to reconstruct realistic environments and robot models, we successfully demonstrate zero-shot Sim2Real transfer, where models trained exclusively on simulation data can perform tasks on a real robot seamlessly, without any fine-tuning. In conclusion, with the unification of these critical components, RealMirror provides a robust framework that significantly accelerates the development of VLA models for humanoid robots. Project page: https://terminators2025.github.io/RealMirror.github.io

cs.RO↗

arXiv2024

Neural Canonical Transformations for Quantum Anharmonic Solids of Lithium

Lithium is a typical quantum solid, characterized by cubic structures at ambient pressure. As the pressure increases, it forms more complex structures and undergoes a metal-to-semiconductor transformation, complicating theoretical and experimental analyses. We employ the neural canonical transformation approach, an \textit{ab initio} variational method based on probabilistic generative models, to investigate the quantum anharmonic effects in lithium solids at finite temperatures. This approach combines a normalizing flow for phonon excited-state wave functions with a probabilistic model for the occupation of energy levels, optimized jointly to minimize the free energy. Our results indicate that quantum anharmonicity lowers the \textit{bcc}-\textit{fcc} transition temperature compared to classical molecular dynamics predictions. At high pressures, the predicted fractional coordinates of lithium atoms in the \textit{cI16} structure show good quantitative agreement with experimental observations. Finally, contrary to previous beliefs, we find that the poor metallic \textit{oC88} structure is stabilized by the potential energy surface obtained via high-accuracy electronic structure calculations, rather than thermal or quantum nuclear effects.

cond-mat.mtrl-sci↗

arXiv2024

LibRPA: A Software Package for Low-scaling First-principles Calculations of Random Phase Approximation Electron Correlation Energy Based on Numerical Atomic Orbitals

LibRPA is a software package designed for efficient calculations of random phase approximation (RPA) electron correlation energies from first principles using numerical atomic orbital (NAOs). Leveraging a localized resolution of identity (LRI) technique, LibRPA achieves $O(N^2)$ or better scaling behavior, making it suitable for large-scale calculation of periodic systems. Implemented in C++ and Python with MPI/OpenMP parallelism, LibRPA integrates seamlessly with NAO-based density functional theory (DFT) packages through flexible file-based and API-based interfaces. In this work, we present the theoretical framework, algorithm, software architecture, and installation and usage guide of LibRPA. Performance benchmarks, including the parallel efficiency with respect to the computational resources and the adsorption energy calculations for H$_2$O molecules on graphene, demonstrate its nearly ideal scalability and numerical reliability. LibRPA offers a useful tool for RPA-based calculations for large-scale extended systems.

cond-mat.mtrl-sci↗

arXiv2023

Sub-quadratic scaling real-space random-phase approximation correlation energy calculations for periodic systems with numerical atomic orbitals

The random phase approximation (RPA) as formulated as an orbital-dependent, fifth-rung functional within the density functional theory (DFT) framework offers a promising approach for calculating the ground-state energies and the derived properties of real materials. Its widespread use to large-size, complex materials is however impeded by the significantly increased computational cost, compared to lower-rung functionals. The standard implementation exhibits an $\mathcal{O}(N^4)$-scaling behavior with respect to system size $N$. In this work, we develop a low-scaling RPA algorithm for periodic systems, based on the numerical atomic orbital (NAO) basis-set framework and a localized variant of the resolution of identity (RI) approximation. The rate-determining step for RPA calculations -- the evaluation of non-interacting response function matrix, is reduced from $\mathcal{O}(N^4)$ to $\mathcal{O}(N^2)$ by just exploiting the sparsity of the RI expansion coefficients, resultant from localized RI (LRI) scheme and the strict locality of NAOs. The computational cost of this step can be further reduced to linear scaling if the decay behavior of the Green's function in real space can be further taken into account. Benchmark calculations against existing $\textbf k$-space based implementation confirms the validity and high numerical precision of the present algorithm and implementation. The new RPA algorithm allows us to readily handle three-dimensional, closely-packed solid state materials with over 1000 atoms. The algorithm and numerical techniques developed in this work also have implications for developing low-scaling algorithms for other correlated methods to be applicable to large-scale extended materials.

physics.comp-ph↗

arXiv2022

First-principle calculations of plasmon excitations in graphene,silicene and germanene

Plasmon excitations in graphene, silicene and germanene are studied using linear-response time dependent density functional theory within the random phase approximation (RPA). Here, we examine both the plasmon dispersion behavior and lifetime of extrinsic and intrinsic plasmons for these three materials. For extrinsic plasmons, we found that their properties are closely related to Landau damping. In the region without single-particle excitation (SPE), the plasmon dispersion shows a \sqrt{q} behavior and the lifetime is infinite at the RPA level, while in the single-particle excitation region, the plasmon dispersion shows a quasilinear behavior and the lifetime is finite. Moreover, for intrinsic plasmons, unlike graphene, the plasmon dispersion behavior of silicene and germanene exhibits a two-peak structure, which can be attributed to the complex and hybridized band structure of these two materials.

physics.comp-ph↗