arXiv subjects

Wang Hong

Publications and source records attributed to Wang Hong.

6 recordsLinked to original sources

arXiv2026

OmniTable: A Unified Wide-Table System for Petabyte-Scale LLM Data Curation and Exploration

Data curation is a critical bottleneck in industrial-grade LLM development, where petabyte-scale unstructured corpora are scattered across hundreds of physical tables, feature engineering relies on manual, table-centric pipeline orchestration, and data lineage is largely absent. We present OmniTable as an architecture blueprint for a unified wide-table layer built on Logical Unification, Physical Separation, targeting petabyte-scale LLM data curation and exploration. OmniTable makes four contributions: (1) a unified wide-table abstraction that consolidates multi-source heterogeneous data and thousands of derived features under a single logical schema via logical-physical mapping; (2) declarative feature lifecycle management that automates dependency resolution, execution planning, operator fusion, and lineage tracking, replacing manual pipeline orchestration with a "declare-and-execute" paradigm; (3) an adaptive execution engine with autonomous governance that achieves stable PB-scale feature backfill through heterogeneous compute routing (CPU/GPU), adaptive tuning, UDF-level fault tolerance, and automated storage layout optimization; and (4) hybrid-accelerated data exploration combining a global ID index, transparent OLAP offloading, and background materialized views to deliver second-level point lookups and filtered exports exceeding 20 TB/hour. In production, OmniTable manages over 35 PB of training data across web, code, PDF, and SFT domains, reducing the human-in-the-loop curation cycle from approximately 14 days to approximately 2.5 days (5.6x over the pre-OmniTable production workflow), with consistent feature versioning, auditable lineage, and minimal manual intervention.

cs.DB↗

arXiv2025

Every Step Evolves: Scaling Reinforcement Learning for Trillion-Scale Thinking Model

We present Ring-1T, the first open-source, state-of-the-art thinking model with a trillion-scale parameter. It features 1 trillion total parameters and activates approximately 50 billion per token. Training such models at a trillion-parameter scale introduces unprecedented challenges, including train-inference misalignment, inefficiencies in rollout processing, and bottlenecks in the RL system. To address these, we pioneer three interconnected innovations: (1) IcePop stabilizes RL training via token-level discrepancy masking and clipping, resolving instability from training-inference mismatches; (2) C3PO++ improves resource utilization for long rollouts under a token budget by dynamically partitioning them, thereby obtaining high time efficiency; and (3) ASystem, a high-performance RL framework designed to overcome the systemic bottlenecks that impede trillion-parameter model training. Ring-1T delivers breakthrough results across critical benchmarks: 93.4 on AIME-2025, 86.72 on HMMT-2025, 2088 on CodeForces, and 55.94 on ARC-AGI-1. Notably, it attains a silver medal-level result on the IMO-2025, underscoring its exceptional reasoning capabilities. By releasing the complete 1T parameter MoE model to the community, we provide the research community with direct access to cutting-edge reasoning capabilities. This contribution marks a significant milestone in democratizing large-scale reasoning intelligence and establishes a new baseline for open-source model performance.

cs.CL↗

arXiv2023

The Nanoplasmonic Purcell Effect in Ultrafast and High-Light-Yield Perovskite Scintillators

The development of X-ray scintillators with ultrahigh light yields and ultrafast response times is a long sought-after goal. In this work, we theoretically predict and experimentally demonstrate a fundamental mechanism that pushes the frontiers of ultrafast X-ray scintillator performance: the use of nanoscale-confined surface plasmon polariton modes to tailor the scintillator response time via the Purcell effect. By incorporating nanoplasmonic materials in scintillator devices, this work predicts over 10-fold enhancement in decay rate and 38% reduction in time resolution even with only a simple planar design. We experimentally demonstrate the nanoplasmonic Purcell effect using perovskite scintillators, enhancing the light yield by over 120% to 88 $\pm$ 11 ph/keV, and the decay rate by over 60% to 2.0 $\pm$ 0.2 ns for the average decay time, and 0.7 $\pm$ 0.1 ns for the ultrafast decay component, in good agreement with the predictions of our theoretical framework. We perform proof-of-concept X-ray imaging experiments using nanoplasmonic scintillators, demonstrating 182% enhancement in the modulation transfer function at 4 line pairs per millimeter spatial frequency. This work highlights the enormous potential of nanoplasmonics in optimizing ultrafast scintillator devices for applications including time-of-flight X-ray imaging and photon-counting computed tomography.

physics.optics↗

arXiv2021

Diffusions and anomalous diffusions of charged particles in the plateau regime of toroidal plasma

The diffusions and anomalous diffusions of charged particles in the plateau regime of toroidal plasma with a Maxwellian velocity distribution are studied. The transport theory of plasma is discussed under the toroidal coordinate system and the magnetic field is assumed to be axisymmetric. Based on the diffusion flux obtained in the toroidal plasma, we derive the diffusion coefficient tensor and thermal diffusion coefficient tensor, where two angular diffusion coefficients are the anomalous diffusions, depending on the radial position and having nothing to do with the magnetic field. All the radial transport coefficients and the normal angular coefficients are inversely proportional to square of the gyrofrequency and so to square of the magnetic induction. The numerical analyses show clearly dependences of these complex transport coefficients on the temperature, the magnetic fluid stability safe factor and the effective collision frequency, respectively.

physics.plasm-ph↗

arXiv2021

The ion acoustic solitary waves in the four component complex plasma with a Cairns-Tsallis distribution

We study the ion acoustic solitary waves in the four-component complex plasma consisting of cold inertial ions, positrons, cold and hot (two-temperature) electrons, where the electrons and positrons are the Cairns-Tsallis distribution and have different nonthermal and nonextensive parameters. Base on the plasma hydrodynamic equations and Sagdeev pseudo-potential theory, we derive the conditions of the solitary waves to exist in the plasma, such as the Sagdeev pseudo-potential, the normalized electrostatic potential, the lower and upper limits of Mach number, and the compressive/rarefactive solitary wave. We show that, according to the present study, the solitary wave solutions exist only if the positrons are a Maxwellian distribution and the two-temperature electrons have the same nonextensive and nonthermal parameters. Numerical analyses are made for the conditions of solitary waves depending on the nonextensive and nonthermal parameters in the Cairns-Tsallis distribution.

physics.plasm-ph↗

arXiv2016

NLO QCD + EW corrections to $ZZZ$ production with subsequent leptonic decays at LHC

In this paper we present the NLO QCD + NLO EW corrections to the $ZZZ$ production with subsequent $Z$-boson leptonic decays at the LHC by adopting the improved narrow width approximation, which takes into account off-shell contributions and spin correlations. Integrated cross sections at $13$, $14$, $33$ and $100~{\rm TeV}$ hadron colliders and various kinematic distributions are presented. Our numerical results show that the jet emission correction accounts for a large part of the total QCD correction, especially in the high energy region. By applying a proper cut to the jet transverse momentum, e.g., $p_{\rm \,T,jet} > p_{\rm \,T,jet}^{\rm \,\,\,cut}=50~{\rm GeV}$, the jet emission correction can be reduced and the cross section is less dependent on the factorization/renormalization scale. This work reveals that both the NLO QCD and the NLO EW corrections are significant. For example, the NLO QCD, NLO EW and NLO QCD + EW relative corrections in the inclusive event selection scheme at the $13~{\rm TeV}$ LHC can reach $63.2\%$, $-9.6\%$ and $47.5\%$, respectively, which means that neither the NLO QCD nor the NLO EW correction is negligible in precision calculations.

hep-ph↗