arXiv subjects

Y. Wu

Publications and source records attributed to Y. Wu.

At least 37 records · Page 2Linked to original sources

arXiv2024

Math-Shepherd: Verify and Reinforce LLMs Step-by-step without Human Annotations

In this paper, we present an innovative process-oriented math process reward model called \textbf{Math-Shepherd}, which assigns a reward score to each step of math problem solutions. The training of Math-Shepherd is achieved using automatically constructed process-wise supervision data, breaking the bottleneck of heavy reliance on manual annotation in existing work. We explore the effectiveness of Math-Shepherd in two scenarios: 1) \textit{Verification}: Math-Shepherd is utilized for reranking multiple outputs generated by Large Language Models (LLMs); 2) \textit{Reinforcement Learning}: Math-Shepherd is employed to reinforce LLMs with step-by-step Proximal Policy Optimization (PPO). With Math-Shepherd, a series of open-source LLMs demonstrates exceptional performance. For instance, the step-by-step PPO with Math-Shepherd significantly improves the accuracy of Mistral-7B (77.9\%$\to$84.1\% on GSM8K and 28.6\%$\to$33.0\% on MATH). The accuracy can be further enhanced to 89.1\% and 43.5\% on GSM8K and MATH with the verification of Math-Shepherd, respectively. We believe that automatic process supervision holds significant potential for the future evolution of LLMs.

cs.AI↗

arXiv2024

DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence

The rapid development of large language models has revolutionized code intelligence in software development. However, the predominance of closed-source models has restricted extensive research and development. To address this, we introduce the DeepSeek-Coder series, a range of open-source code models with sizes from 1.3B to 33B, trained from scratch on 2 trillion tokens. These models are pre-trained on a high-quality project-level code corpus and employ a fill-in-the-blank task with a 16K window to enhance code generation and infilling. Our extensive evaluations demonstrate that DeepSeek-Coder not only achieves state-of-the-art performance among open-source code models across multiple benchmarks but also surpasses existing closed-source models like Codex and GPT-3.5. Furthermore, DeepSeek-Coder models are under a permissive license that allows for both research and unrestricted commercial use.

cs.SE↗

arXiv2024

DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models

In the era of large language models, Mixture-of-Experts (MoE) is a promising architecture for managing computational costs when scaling up model parameters. However, conventional MoE architectures like GShard, which activate the top-$K$ out of $N$ experts, face challenges in ensuring expert specialization, i.e. each expert acquires non-overlapping and focused knowledge. In response, we propose the DeepSeekMoE architecture towards ultimate expert specialization. It involves two principal strategies: (1) finely segmenting the experts into $mN$ ones and activating $mK$ from them, allowing for a more flexible combination of activated experts; (2) isolating $K_s$ experts as shared ones, aiming at capturing common knowledge and mitigating redundancy in routed experts. Starting from a modest scale with 2B parameters, we demonstrate that DeepSeekMoE 2B achieves comparable performance with GShard 2.9B, which has 1.5 times the expert parameters and computation. In addition, DeepSeekMoE 2B nearly approaches the performance of its dense counterpart with the same number of total parameters, which set the upper bound of MoE models. Subsequently, we scale up DeepSeekMoE to 16B parameters and show that it achieves comparable performance with LLaMA2 7B, with only about 40% of computations. Further, our preliminary efforts to scale up DeepSeekMoE to 145B parameters consistently validate its substantial advantages over the GShard architecture, and show its performance comparable with DeepSeek 67B, using only 28.5% (maybe even 18.2%) of computations.

cs.CL↗

arXiv2024

DeepSeek LLM: Scaling Open-Source Language Models with Longtermism

The rapid development of open-source large language models (LLMs) has been truly remarkable. However, the scaling law described in previous literature presents varying conclusions, which casts a dark cloud over scaling LLMs. We delve into the study of scaling laws and present our distinctive findings that facilitate scaling of large scale models in two commonly used open-source configurations, 7B and 67B. Guided by the scaling laws, we introduce DeepSeek LLM, a project dedicated to advancing open-source language models with a long-term perspective. To support the pre-training phase, we have developed a dataset that currently consists of 2 trillion tokens and is continuously expanding. We further conduct supervised fine-tuning (SFT) and Direct Preference Optimization (DPO) on DeepSeek LLM Base models, resulting in the creation of DeepSeek Chat models. Our evaluation results demonstrate that DeepSeek LLM 67B surpasses LLaMA-2 70B on various benchmarks, particularly in the domains of code, mathematics, and reasoning. Furthermore, open-ended evaluations reveal that DeepSeek LLM 67B Chat exhibits superior performance compared to GPT-3.5.

cs.CL↗

arXiv2023

Possible manifestation of topological superconductivity and Majorana bound states in the microwave response of thin FeSe1-xTex film

The paper analyzes the characteristics of the microwave (MW) response of FeSe1-xTex films based on the results of measuring the impedance properties of the films in the X-band for two orientations of the film in the MW magnetic field, perpendicular and parallel. The analysis of the temperature dependence of the microwave response of a film with a perpendicular orientation (in which the peculiarity of the response is manifested) was carried out by means of physical considerations, taking into account also the results of the research of this superconductor by other authors using the ARPES technique and tunneling spectroscopy. It was concluded that with perpendicular orientation, two competing mechanisms of MW energy dissipation in the film can occur, one of which leads to the increase in energy dissipation caused by magnetic vortices with an MW field, and the other to its decrease due to the emergence of Majorana bound states with zero energy

cond-mat.supr-con↗

arXiv2023

Nernst Sign-Reversal in the Hexatic Vortex Phase of Weakly Disordered a-MoGe Thin Films

The hexatic phase is an intermediate stage in the melting process of a 2D crystal due to topological defects. Recently, this exotic phase was experimentally identified in the vortex lattice of 2D weakly disordered superconducting MoGe by scanning tunneling microscopic measurements. Here we study this vortex state by the Nernst effect, which is an effective and sensitive tool to detect vortex motion, especially in the superconducting fluctuation regime. We find a surprising Nernst sign reversal at the melting transition of the hexatic phase. We propose that they are a consequence of vortex dislocations in the hexatic state which diffuse preferably from the cold to hot.

cond-mat.supr-con↗

arXiv2023

Design and Testing of Cesium Atomic Concentration Detection System Based on TDLAS

In order to better build the Neutral Beam Injector with Negative Ion Source (NNBI), the pre-research on key technologies has been carried out for the Comprehensive Research Facility for Fusion Technology (CRAFT). Cesium seeding into negative-ion sources is a prerequisite to obtain the required negative hydrogen ion. The performance of ion source largely depends on the cesium conditions in the source. It is very necessary to quantitatively measure the amount of cesium in the source during the plasma on and off periods (vacuum stage). This article uses the absorption peak of cesium atoms near 852.1nm to build a cesium atom concentration detection system based on Tunable Diode Laser Absorption Spectroscopy (TDLAS) technology. The test experiment based on the cesium cell is carried out, obtained the variation curve of cesium concentration at different temperatures. The experimental results indicate that: the system detection range is within 5*10E6-2.5*10E7 pieces/cm3 and the system resolution better than 1*10E6 pieces/cm3.

physics.ins-det↗

arXiv2023

Probing strong-field QED in beam-plasma collisions

Ongoing progress in laser and accelerator technology opens new possibilities in high-field science, notably to investigate the largely unexplored strong-field quantum electrodynamics (SFQED) regime where electron-positron pairs can be created directly from light-matter or even light-vacuum interactions. Laserless strategies such as beam-beam collisions have also been proposed to access the nonperturbative limit of SFQED. Here we report on a concept to probe SFQED by harnessing the interaction between a high-charge, ultrarelativistic electron beam and a solid conducting target. When impinging onto the target surface, the beam self fields are reflected, partly or fully, depending on the beam shape; in the rest frame of the beam electrons, these fields can exceed the Schwinger field, thus triggering SFQED effects such as quantum nonlinear inverse Compton scattering and nonlinear Breit-Wheeler electron-positron pair creation. Through reduced modeling and kinetic numerical simulations, we show that this single-beam setup can achieve interaction conditions similar to those envisioned in beam-beam collisions, but in a simpler and more controllable way owing to the automatic overlap of the beam and driving fields. This scheme thus eases the way to precision studies of SFQED and is also a promising milestone towards laserless studies of nonperturbative SFQED.

physics.plasm-ph↗

arXiv2023

Absence of $3a_0$ Charge Density Wave Order in the Infinite Layer Nickelates

A hallmark of many unconventional superconductors is the presence of many-body interactions which give rise to broken symmetry states intertwined with superconductivity. Recent resonant soft x-ray scattering experiments report commensurate $3a_0$ charge density wave order in the infinite layer nickelates, which has important implications regarding the universal interplay between charge order and superconductivity in both the cuprates and nickelates. Here, we present x-ray scattering and spectroscopy measurements on a series of NdNiO$_{2+x}$ samples which reveal that the signatures of charge density wave order are absent in fully reduced, single-phase NdNiO$_2$. The $3a_0$ superlattice peak instead originates from a partially reduced impurity phase where excess apical oxygens form ordered rows with 3 unit cell periodicity. The absence of any observable charge density wave order in NdNiO$_2$ highlights a crucial difference between the phase diagrams of the cuprate and nickelate superconductors.

cond-mat.supr-con↗

arXiv2023

Multi-k magnetic structure and large anomalous Hall effect in candidate magnetic Weyl semimetal NdAlGe

The magnetic structure, magnetoresistance, and Hall effect of non-centrosymmetric magnetic semimetal NdAlGe are investigated revealing an unusual magnetic state and anomalous transport properties that are associated with the electronic structure of this non-centrosymmetric compound. The magnetization and magnetoresistance measurements are both highly anisotropic and indicate an Ising-like magnetic system. The magnetic structure is complex in that it involves three magnetic ordering vectors including an incommensurate spin density wave and commensurate ferrimagnetic state in zero field. We have discovered a large anomalous Hall conductivity that reaches = 430 Ω-1cm-1 implying that it originates from an intrinsic Berry curvature effect stemming from Weyl nodes found in the electronic structure. These electronic structure calculations indicate the presence of nested Fermi surface pockets with nesting wave vectors similar to the measured magnetic ordering wavevector and the presence of Weyl nodes in proximity to the Fermi surface. We associate the incommensurate magnetic structure with the large anomalous Hall response to be the result of the combination of Fermi surface nesting and the Berry curvature associated with Weyl nodes.

cond-mat.mtrl-sci↗

arXiv2023

Improved Calibration of RF Cavities for Relativistic Electron Beams: Effects of Secondary Corrections and Experimental Verification

In the aspect of longitudinal beam bunching, the bunching strength can be controlled by the RF cavity phase and voltage. However, these machine parameters are different from those that interact with the beam itself. In order to gain control of the beam-cavity interaction, cavity calibration must be performed. Furthermore, it relies on fitting the beam energy gain versus cavity phase to a calibration function. Under the conventional assumption of relativistic beam conditions, the calibration function is a first harmonic sinusoidal function (a sinusoidal function with a period of 2π). However, this expression is insufficient for a high-voltage bunching cavity. Due to beam acceleration inside the cavity, an energy bias and a second harmonic function should be included to modify the conventional calibration function, even for a relativistic electron beam. In this paper, we will derive this modification and provide a comparison to both the Coherent Electron Cooling Experiment and the IMPACT-T simulation, respectively.

physics.acc-ph↗

arXiv2022

Understanding the Anomalous Hall effect in Co$_{1/3}$NbS$_{2}$ from crystal and magnetic structures

A large anomalous Hall effect (AHE) has recently been observed in the intercalated transition metal dichalcogenide (TMDC) Co$_{1/3}$NbS$_{2}$ below a known magnetic phase transition at $T_N$ = 29 K. The spins in this material are widely believed to order in a highly symmetric collinear antiferromagnetic configuration, causing extensive debate about how reports of an AHE can be reconciled with such a state. In this article, we address this controversy by presenting new neutron diffraction data on single crystals of Co$_{1/3}$NbS$_{2}$ and an analysis that implies that moments in this material order into a non-collinear configuration, but one that maintains the same refelction symmetries as the collinear phase. We present new transport and magneto-optic Kerr measurements which show that AHE signatures persist below $T_N$ to temperatures as low as $T$ = 5 K and firmly associate them with the long-range antiferromagnetic order. Finally, we show that these AHE signatures can be quantitatively reproduced by density functional theory (DFT) calculations based on the lattice and spin state determined with neutron diffraction. These combined findings establishes the veracity of the 'crystal Hall effect' picture, which shows how such effects can emerge from the shape of magnetic orbitals in compounds containing chiral lattice symmetry regardless of the symmetry of the ordered spin configuration. These results illuminate a new path for the discovery of anomalous Hall materials and motivate a targeted study of the transport properties of intercalated TMDCs and other compounds containing antiferromagnetic order and chiral lattice symmetry.

cond-mat.mtrl-sci↗

arXiv2022

Report of the Topical Group on Higgs Physics for Snowmass 2021: The Case for Precision Higgs Physics

A future Higgs Factory will provide improved precision on measurements of Higgs couplings beyond those obtained by the LHC, and will enable a broad range of investigations across the fields of fundamental physics, including the mechanism of electroweak symmetry breaking, the origin of the masses and mixing of fundamental particles, the predominance of matter over antimatter, and the nature of dark matter. Future colliders will measure Higgs couplings to a few per cent, giving a window to beyond the Standard Model (BSM) physics in the 1-10 TeV range. In addition, they will make precise measurements of the Higgs width, and characterize the Higgs self-coupling. This report details the work of the EF01 and EF02 working groups for the Snowmass 2021 study.

hep-ph↗

arXiv2022

X-ray morphology due to charge-exchange emissions used to study the global structure around Mars

Soft x-ray emissions induced by solar wind ions that collide with neutral material in the solar system have been detected around planets, and were proposed as a remote probe for the solar wind interaction with the Martian exosphere. A multi-fluid three-dimensional magneto-hydrodynamic model is adopted to derive the global distributions of solar wind particles. Spherically symmetric exospheric H, H$_2$, He, O, and CO$_2$ density profiles and a sophisticated hybrid model that includes charge-exchange and proton/neutral excitation processes are used to study the low triplet line ratio $G=\frac{i+f}{r}$ (0.77$\pm$0.58) of O VII and total x-ray luminosity around Mars. We further calculate the emission factor $α$-value with different neutrals over a wide ion abundance and velocity ranges. Our results are in good agreement with those of previous reports. The evolution of the charge stage of solar wind ions shows that sequential recombination due to charge-exchange can be negligible at the interaction region. This only appears below the altitude of 400~km. The anonymous low disk $G$ ratio can be easily explained by the collisional quenching effect at neutral densities higher than 10$^{11}$cm$^{-3}$. However, the quenching contribution is small in Mars' exosphere and only appears below 400~km. Charge-exchange with H$_2$ and N$_2$ is still the most likely reason for this low $G$-ratio. X-ray emissivity maps in collisions with different neutrals differ from each other. A clear bow shock in the collision with all the neutrals is in accordance with previous reports. The resulting total x-ray luminosity of 6.55~MW shows a better agreement with the XMM-Newton observation of 12.8$\pm$1.4~MW than that of previous predictions.

astro-ph.EP↗

arXiv2022

A scaling relation in [C II]-detected galaxies and its likely application in cosmology

We identify and investigate a possible correlation between the $\rm{[CII]} 158μm$ luminosity and linewidth in the $\rm{[CII]}$-detected galaxies. Observationally, the strength of the $\rm{[CII]} 158μm$ emission line is usually stronger than that of the CO emission line and this $\rm{[CII]}$ line has been used as another tracer of the galactic characteristics. Moreover, many $\rm{[CII]}$-detected galaxies are identified in $z > 4$. Motivated by previous studies of the CO luminosity - FWHM correlation relation (LFR) and the available new $\rm{[CII]}$ measurements, we compile samples of the $\rm{[CII]}$-detected galaxies in the literature and perform the linear regression analysis. The $\rm{[CII]}$ LFR is confirmed at a robust level. We also demonstrate the possible application of the $\rm{[CII]}$ LFR by utilizing it on the distance measurement of the high-$z$ galaxy. As a result, we extend the cosmic spatial scale beyond the redshift $z$ of $7$. With the outcome of the distance measurement, we constrain the cosmology parameters in the Chevallier-Polarski-Linder model, which considers the evolution of dark energy. Consequently, the uncertainties of the $\textit{w}_{0}$ and $\textit{w}_{a}$ are reduced significantly when the measured distance data of the $\rm{[CII]}$-detected galaxies are included in the cosmological parameter constraint, exemplifying the potential of using the $\rm{[CII]}$-detected galaxies as a tracer to constrain the cosmological parameters.

astro-ph.GA↗

arXiv2022

Ferroelectric Solitons Crafted in Epitaxial Bismuth Ferrite Superlattices

In ferroelectrics, complex interactions among various degrees of freedom enable the condensation of topologically protected polarization textures. Known as ferroelectric solitons, these particle-like structures represent a new class of materials with promise for beyond CMOS technologies due to their ultrafine size and sensitivity to external stimuli. Such polarization textures have scarcely been reported in multiferroics. Here, we report a range of soliton topologies in bismuth ferrite strontium titanate superlattices. High-resolution piezoresponse force microscopy and Cs-corrected high-angle annular dark-field scanning transmission electron microscopy reveal a zoo of topologies, and polarization displacement mapping of planar specimens reveals center-convergent and divergent topological defects as small as 3 nm. Phase field simulations verify that some of these topologies can be classed as bimerons, with a topological charge of plus and minus one, and first-principles-based effective Hamiltonian computations show that the co-existence of such structures can lead to non-integer topological charges, a first observation in a BiFeO3-based system. Our results open new opportunities in multiferroic topotronics.

cond-mat.mtrl-sci↗

arXiv2022

Positive solutions of the Gross-Pitaevskii equation for energy critical and supercritical nonlinearities

We consider positive and spatially decaying solutions to the Gross-Pitaevskii equation with a harmonic potential. For the energy-critical case, there exists a ground state if and only if the frequency belongs to (1,3) in three dimensions and in (0,d) in d dimensions. We give a precise description on asymptotic behaviors of the ground state up to the leading order term for different values of d. For the energy-supercritical case, there exists a singular solution for some frequency in (0,d). We compute the Morse index of the singular solution in the class of radial functions and show that the Morse index is infinite in the oscillatory case, is equal to 1 or 2 in the monotone case for nonlinearity powers not large enough and is equal to 1 in the monotone case for nonlinearity power sufficiently large.

math.AP↗

arXiv2022

A Generative deep learning approach for shape recognition of arbitrary objects from phaseless acoustic scattering data

We propose and demonstrate a generative deep learning approach for the shape recognition of an arbitrary object from its acoustic scattering properties. The strategy exploits deep neural networks to learn the mapping between the latent space of a two-dimensional acoustic object and the far-field scattering amplitudes. A neural network is designed as an Adversarial autoencoder and trained via unsupervised learning to determine the latent space of the acoustic object. Important structural features of the object are embedded in lower-dimensional latent space which supports the modeling of a shape generator and accelerates the learning in the inverse design process.The proposed inverse design uses the variational inference approach with encoder and decoder-like architecture where the decoder is composed of two pretrained neural networks, the generator and the forward model. The data-driven framework finds an accurate solution to the ill-posed inverse scattering problem, where non-unique solution space is overcome by the multifrequency phaseless far-field patterns. This inverse method is a powerful design tool that does not require complex analytical calculation and opens up new avenues for practical realization, automatic recognition of arbitrary shaped submarines or large fish, and other underwater applications.

cs.SD↗