arXiv subjects

Peng Cheng

Publications and source records attributed to Peng Cheng.

At least 163 records · Page 9Linked to original sources

arXiv2024

SuperBench: Improving Cloud AI Infrastructure Reliability with Proactive Validation

Reliability in cloud AI infrastructure is crucial for cloud service providers, prompting the widespread use of hardware redundancies. However, these redundancies can inadvertently lead to hidden degradation, so called "gray failure", for AI workloads, significantly affecting end-to-end performance and concealing performance issues, which complicates root cause analysis for failures and regressions. We introduce SuperBench, a proactive validation system for AI infrastructure that mitigates hidden degradation caused by hardware redundancies and enhances overall reliability. SuperBench features a comprehensive benchmark suite, capable of evaluating individual hardware components and representing most real AI workloads. It comprises a Validator which learns benchmark criteria to clearly pinpoint defective components. Additionally, SuperBench incorporates a Selector to balance validation time and issue-related penalties, enabling optimal timing for validation execution with a tailored subset of benchmarks. Through testbed evaluation and simulation, we demonstrate that SuperBench can increase the mean time between incidents by up to 22.61x. SuperBench has been successfully deployed in Azure production, validating hundreds of thousands of GPUs over the last two years.

cs.DC↗

arXiv2024

NewsBench: A Systematic Evaluation Framework for Assessing Editorial Capabilities of Large Language Models in Chinese Journalism

We present NewsBench, a novel evaluation framework to systematically assess the capabilities of Large Language Models (LLMs) for editorial capabilities in Chinese journalism. Our constructed benchmark dataset is focused on four facets of writing proficiency and six facets of safety adherence, and it comprises manually and carefully designed 1,267 test samples in the types of multiple choice questions and short answer questions for five editorial tasks in 24 news domains. To measure performances, we propose different GPT-4 based automatic evaluation protocols to assess LLM generations for short answer questions in terms of writing proficiency and safety adherence, and both are validated by the high correlations with human evaluations. Based on the systematic evaluation framework, we conduct a comprehensive analysis of ten popular LLMs which can handle Chinese. The experimental results highlight GPT-4 and ERNIE Bot as top performers, yet reveal a relative deficiency in journalistic safety adherence in creative writing tasks. Our findings also underscore the need for enhanced ethical guidance in machine-generated journalistic content, marking a step forward in aligning LLMs with journalistic standards and safety considerations.

cs.CL↗

arXiv2024

Skywork-MoE: A Deep Dive into Training Techniques for Mixture-of-Experts Language Models

In this technical report, we introduce the training methodologies implemented in the development of Skywork-MoE, a high-performance mixture-of-experts (MoE) large language model (LLM) with 146 billion parameters and 16 experts. It is initialized from the pre-existing dense checkpoints of our Skywork-13B model. We explore the comparative effectiveness of upcycling versus training from scratch initializations. Our findings suggest that the choice between these two approaches should consider both the performance of the existing dense checkpoints and the MoE training budget. We highlight two innovative techniques: gating logit normalization, which improves expert diversification, and adaptive auxiliary loss coefficients, allowing for layer-specific adjustment of auxiliary loss coefficients. Our experimental results validate the effectiveness of these methods. Leveraging these techniques and insights, we trained our upcycled Skywork-MoE on a condensed subset of our SkyPile corpus. The evaluation results demonstrate that our model delivers strong performance across a wide range of benchmarks.

cs.CL↗

arXiv2024

LongSkywork: A Training Recipe for Efficiently Extending Context Length in Large Language Models

We introduce LongSkywork, a long-context Large Language Model (LLM) capable of processing up to 200,000 tokens. We provide a training recipe for efficiently extending context length of LLMs. We identify that the critical element in enhancing long-context processing capability is to incorporate a long-context SFT stage following the standard SFT stage. A mere 200 iterations can convert the standard SFT model into a long-context model. To reduce the effort in collecting and annotating data for long-context language modeling, we develop two novel methods for creating synthetic data. These methods are applied during the continual pretraining phase as well as the Supervised Fine-Tuning (SFT) phase, greatly enhancing the training efficiency of our long-context LLMs. Our findings suggest that synthetic long-context SFT data can surpass the performance of data curated by humans to some extent. LongSkywork achieves outstanding performance on a variety of long-context benchmarks. In the Needle test, a benchmark for long-context information retrieval, our models achieved perfect accuracy across multiple context spans. Moreover, in realistic application scenarios, LongSkywork-13B demonstrates performance on par with Claude2.1, the leading long-context model, underscoring the effectiveness of our proposed methods.

cs.CL↗

arXiv2024

UHGEval: Benchmarking the Hallucination of Chinese Large Language Models via Unconstrained Generation

Large language models (LLMs) have emerged as pivotal contributors in contemporary natural language processing and are increasingly being applied across a diverse range of industries. However, these large-scale probabilistic statistical models cannot currently ensure the requisite quality in professional content generation. These models often produce hallucinated text, compromising their practical utility in professional contexts. To assess the authentic reliability of LLMs in text generation, numerous initiatives have developed benchmark evaluations for hallucination phenomena. Nevertheless, these benchmarks frequently utilize constrained generation techniques due to cost and temporal constraints. These techniques encompass the use of directed hallucination induction and strategies that deliberately alter authentic text to produce hallucinations. These approaches are not congruent with the unrestricted text generation demanded by real-world applications. Furthermore, a well-established Chinese-language dataset dedicated to the evaluation of hallucinations in text generation is presently lacking. Consequently, we have developed an Unconstrained Hallucination Generation Evaluation (UHGEval) benchmark, designed to compile outputs produced with minimal restrictions by LLMs. Concurrently, we have established a comprehensive benchmark evaluation framework to aid subsequent researchers in undertaking scalable and reproducible experiments. We have also executed extensive experiments, evaluating prominent Chinese language models and the GPT series models to derive professional performance insights regarding hallucination challenges.

cs.CL↗

arXiv2024

Evidence for Multiferroicity in Single-Layer CuCrSe$_2$

Multiferroic materials, which simultaneously exhibit ferroelectricity and magnetism, have attracted substantial attention due to their fascinating physical properties and potential technological applications. With the trends towards device miniaturization, there is an increasing demand for the persistence of multiferroicity in single-layer materials at elevated temperatures. Here, we report high-temperature multiferroicity in single-layer CuCrSe$_2$, which hosts room-temperature ferroelectricity and 120 K ferromagnetism. Notably, the ferromagnetic coupling in single-layer CuCrSe$_2$ is enhanced by the ferroelectricity-induced orbital shift of Cr atoms, which is distinct from both types I and II multiferroicity. These findings are supported by a combination of second-harmonic generation, piezo-response force microscopy, scanning transmission electron microscopy, magnetic, and Hall measurements. Our research provides not only an exemplary platform for delving into intrinsic magnetoelectric interactions at the single-layer limit but also sheds light on potential development of electronic and spintronic devices utilizing two-dimensional multiferroics.

cond-mat.mtrl-sci↗

arXiv2024

Notes on Wormhole Cancellation and Factorization

In AdS/CFT, partition functions of decoupled CFTs living on separate asymptotic boundaries factorize. However, the presence of bulk wormholes connecting different boundaries tends to spoil the factorization of the bulk partition function, which leads to a disagreement between the two sides. In this paper, we present two examples where wormhole contributions cancel each other in bulk partition function calculations, thus the bulk factorization can be realized. The first example is in 2-dimensional Jackiw-Teitelboim (JT) gravity, where the proposed way of realizing the cancellation resides in the extra complex phases associated with different wormholes. The phases arise due to the degenerate vacua structure. In the example of the Sachdev-Ye-Kitaev (SYK) model, the cancellation can be achieved due to the distribution of the wormhole saddles on a complex plane. The two examples demonstrate a way of realizing bulk partition function factorization by extending the Hilbert space and dressing wormhole saddles with extra phases.

hep-th↗

arXiv2024

Reasoning on Efficient Knowledge Paths:Knowledge Graph Guides Large Language Model for Domain Question Answering

Large language models (LLMs), such as GPT3.5, GPT4 and LLAMA2 perform surprisingly well and outperform human experts on many tasks. However, in many domain-specific evaluations, these LLMs often suffer from hallucination problems due to insufficient training of relevant corpus. Furthermore, fine-tuning large models may face problems such as the LLMs are not open source or the construction of high-quality domain instruction is difficult. Therefore, structured knowledge databases such as knowledge graph can better provide domain background knowledge for LLMs and make full use of the reasoning and analysis capabilities of LLMs. In some previous works, LLM was called multiple times to determine whether the current triplet was suitable for inclusion in the subgraph when retrieving subgraphs through a question. Especially for the question that require a multi-hop reasoning path, frequent calls to LLM will consume a lot of computing power. Moreover, when choosing the reasoning path, LLM will be called once for each step, and if one of the steps is selected incorrectly, it will lead to the accumulation of errors in the following steps. In this paper, we integrated and optimized a pipeline for selecting reasoning paths from KG based on LLM, which can reduce the dependency on LLM. In addition, we propose a simple and effective subgraph retrieval method based on chain of thought (CoT) and page rank which can returns the paths most likely to contain the answer. We conduct experiments on three datasets: GenMedGPT-5k [14], WebQuestions [2], and CMCQA [21]. Finally, RoK can demonstrate that using fewer LLM calls can achieve the same results as previous SOTAs models.

cs.CL↗

arXiv2024

Site-ordering/disordering-induced magnetic textures in a vdW ferromagnet by competing global and broken inversion-symmetry

Fe5GeTe2 single crystals can be divided into nonquenched (NQ) and quench-cooled (QC) phases with different magnetic properties. A comprehensive understanding of the magnetic property variations in the NQ and QC phases is imperative for guiding Fe5GeTe2 towards spintronics applications; however, it remains elusive. Here, we report a real-space study on the structural and magnetic properties of these two magnetic phases using cryogenic magnetic force microscopy and scanning tunneling microscopy. The thermal history introduces disorder and order to the Fe(1) sites, resulting in the NQ and QC phases exhibiting global and broken inversion symmetry, respectively. The observed magnetic domain transitions (branching to labyrinthine) in the spin reorientation process and the distinct 3D spin textures stabilized by magnetic dipolar interaction observed in field-dependent studies allow the NQ phase to exhibit a more resilient global magnetic state. In contrast, the QC phase exhibits enhanced magnetic anisotropy, resulting in a higher TC. Meanwhile, the Dzyaloshinskii-Moriya interaction (DMI) introduced by the broken inversion symmetry causes the QC phase to exhibit a localized magnetic state: no domain transformation occurs during spin reorientation, and irregular domain states are observed in field-related studies. Our work provides an important reference for understanding the complex magnetic properties in Fe5GeTe2.

cond-mat.mtrl-sci↗

arXiv2024

Wait to be Faster: a Smart Pooling Framework for Dynamic Ridesharing

Ridesharing services, such as Uber or Didi, have attracted considerable attention in recent years due to their positive impact on environmental protection and the economy. Existing studies require quick responses to orders, which lack the flexibility to accommodate longer wait times for better grouping opportunities. In this paper, we address a NP-hard ridesharing problem, called Minimal Extra Time RideSharing (METRS), which balances waiting time and group quality (i.e., detour time) to improve riders' satisfaction. To tackle this problem, we propose a novel approach called WATTER (WAit To be fasTER), which leverages an order pooling management algorithm allowing orders to wait until they can be matched with suitable groups. The key challenge is to customize the extra time threshold for each order by reducing the original optimization objective into a convex function of threshold, thus offering a theoretical guarantee to be optimized efficiently. We model the dispatch process using a Markov Decision Process (MDP) with a carefully designed value function to learn the threshold. Through extensive experiments on three real datasets, we demonstrate the efficiency and effectiveness of our proposed approaches.

cs.DB↗

arXiv2024

Enhancing Cyber-Resiliency of DER-based SmartGrid: A Survey

The rapid development of information and communications technology has enabled the use of digital-controlled and software-driven distributed energy resources (DERs) to improve the flexibility and efficiency of power supply, and support grid operations. However, this evolution also exposes geographically-dispersed DERs to cyber threats, including hardware and software vulnerabilities, communication issues, and personnel errors, etc. Therefore, enhancing the cyber-resiliency of DER-based smart grid - the ability to survive successful cyber intrusions - is becoming increasingly vital and has garnered significant attention from both industry and academia. In this survey, we aim to provide a systematical and comprehensive review regarding the cyber-resiliency enhancement (CRE) of DER-based smart grid. Firstly, an integrated threat modeling method is tailored for the hierarchical DER-based smart grid with special emphasis on vulnerability identification and impact analysis. Then, the defense-in-depth strategies encompassing prevention, detection, mitigation, and recovery are comprehensively surveyed, systematically classified, and rigorously compared. A CRE framework is subsequently proposed to incorporate the five key resiliency enablers. Finally, challenges and future directions are discussed in details. The overall aim of this survey is to demonstrate the development trend of CRE methods and motivate further efforts to improve the cyber-resiliency of DER-based smart grid.

eess.SY↗

arXiv2024

Layer-dependent Raman spectroscopy of ultrathin Ta$_2$Pd$_3$Te$_5$

Two-dimensional topological insulators (2DTIs) or quantum spin Hall insulators are attracting increasing attention due to their potential applications in next-generation spintronic devices. Despite their promising prospects, realizable 2DTIs are still limited. Recently, Ta2Pd3Te5, a semiconducting van der Waals material, has shown spectroscopic evidence of quantum spin Hall states. However, achieving controlled preparation of few- to monolayer samples, a crucial step in realizing quantum spin Hall devices, has not yet been achieved. In this work, we fabricated few- to monolayer Ta$_2$Pd$_3$Te$_5$ and performed systematic thickness- and temperature-dependent Raman spectroscopy measurements. Our results demonstrate that Raman spectra can provide valuable information to determine the thickness of Ta2Pd3Te5 thin flakes. Moreover, our angle-resolved polarized Raman (ARPR) spectroscopy measurements show that the intensities of the Raman peaks are strongly anisotropic due to the quasi-one-dimensional atomic structure, providing a straightforward method to determine its crystalline orientation. Our findings may stimulate further efforts to realize quantum devices based on few or monolayer Ta$_2$Pd$_3$Te$_5$.

cond-mat.mtrl-sci↗

arXiv2024

Baryon Spin and Emergent Hadron Mass

This thesis describes the use of Dyson-Schwinger equations (DSEs) to study baryon bound states in QCD. In this work, octet baryon axial form factors are calculated using a symmetry-preserving treatment of a vector $\times$ vector contact interaction (SCI). The baryons are considered as quark-plus-interacting-diquark bound states, whose structure (wave function) is obtained by solving a Poincaré-covariant Faddeev equation. Since it preserves symmetries, all consequences of partial conservation of the axial current are manifest. For instance, one finds that octet baryon axial properties are consistent with only minor violations of SU(3)-flavour symmetry, being interpreted as a dynamical consequence of emergent hadron mass (EHM). Considering neutral axial currents, the SCI delivers predictions for the flavour separation of octet baryon axial charges and, consequently, produces values for the associated SU(3) singlet, triplet, and octet axial charges. The results indicate that, at the hadron scale $ζ_{\mathcal{H}}$, valence degrees of freedom carry approximately $50\%$ of an octet baryon's total spin. Proton structure is one of the principal topics in hadron physics. Its study is expected to reveal key features of both the origin of mass and strong interaction dynamics. This work therefore extended the above analyses to an examination of in-proton parton helicity (spin) distribution functions (DFs). Using Ansätze for hadron-scale proton polarised valence quark DFs, predictions are delivered for all proton polarised DFs at the measurement scale $ζ_{\rm C}^2 = 3\,$GeV$^2$. The pointwise behaviour of the predicted DFs and, consequently, their moments, shows good agreement with results inferred from data. Based on these results, one finds that experimental measurements of the proton flavour-singlet axial charge should return a value $a_0^{\rm E}(ζ_{\rm C}) = 0.35(2)$.

hep-ph↗

arXiv2024

Phoneme-Based Proactive Anti-Eavesdropping with Controlled Recording Privilege

The widespread smart devices raise people's concerns of being eavesdropped on. To enhance voice privacy, recent studies exploit the nonlinearity in microphone to jam audio recorders with inaudible ultrasound. However, existing solutions solely rely on energetic masking. Their simple-form noise leads to several problems, such as high energy requirements and being easily removed by speech enhancement techniques. Besides, most of these solutions do not support authorized recording, which restricts their usage scenarios. In this paper, we design an efficient yet robust system that can jam microphones while preserving authorized recording. Specifically, we propose a novel phoneme-based noise with the idea of informational masking, which can distract both machines and humans and is resistant to denoising techniques. Besides, we optimize the noise transmission strategy for broader coverage and implement a hardware prototype of our system. Experimental results show that our system can reduce the recognition accuracy of recordings to below 50\% under all tested speech recognition systems, which is much better than existing solutions.

cs.CR↗

arXiv2023

Vulnerability of Machine Learning Approaches Applied in IoT-based Smart Grid: A Review

Machine learning (ML) sees an increasing prevalence of being used in the internet-of-things (IoT)-based smart grid. However, the trustworthiness of ML is a severe issue that must be addressed to accommodate the trend of ML-based smart grid applications (MLsgAPPs). The adversarial distortion injected into the power signal will greatly affect the system's normal control and operation. Therefore, it is imperative to conduct vulnerability assessment for MLsgAPPs applied in the context of safety-critical power systems. In this paper, we provide a comprehensive review of the recent progress in designing attack and defense methods for MLsgAPPs. Unlike the traditional survey about ML security, this is the first review work about the security of MLsgAPPs that focuses on the characteristics of power systems. We first highlight the specifics for constructing the adversarial attacks on MLsgAPPs. Then, the vulnerability of MLsgAPP is analyzed from both the aspects of the power system and ML model. Afterward, a comprehensive survey is conducted to review and compare existing studies about the adversarial attacks on MLsgAPPs in scenarios of generation, transmission, distribution, and consumption, and the countermeasures are reviewed according to the attacks that they defend against. Finally, the future research directions are discussed on the attacker's and defender's side, respectively. We also analyze the potential vulnerability of large language model-based (e.g., ChatGPT) power system applications. Overall, we encourage more researchers to contribute to investigating the adversarial issues of MLsgAPPs.

cs.CR↗

arXiv2023

FP8-LM: Training FP8 Large Language Models

In this paper, we explore FP8 low-bit data formats for efficient training of large language models (LLMs). Our key insight is that most variables, such as gradients and optimizer states, in LLM training can employ low-precision data formats without compromising model accuracy and requiring no changes to hyper-parameters. Specifically, we propose a new FP8 automatic mixed-precision framework for training LLMs. This framework offers three levels of FP8 utilization to streamline mixed-precision and distributed parallel training for LLMs. It gradually incorporates 8-bit gradients, optimizer states, and distributed learning in an incremental manner. Experiment results show that, during the training of GPT-175B model on H100 GPU platform, our FP8 mixed-precision training framework not only achieved a remarkable 39% reduction in real memory usage but also ran 75% faster than the widely adopted BF16 framework (i.e., Megatron-LM), surpassing the speed of Nvidia Transformer Engine by 37%. This largely reduces the training costs for large foundation models. Furthermore, our FP8 mixed-precision training methodology is generic. It can be seamlessly applied to other tasks such as LLM instruction tuning and reinforcement learning with human feedback, offering savings in fine-tuning expenses. Our FP8 low-precision training framework is open-sourced at {https://github.com/Azure/MS-AMP}{aka.ms/MS.AMP}.

cs.LG↗

arXiv2023

Masked Diffusion Models Are Fast Distribution Learners

Diffusion model has emerged as the \emph{de-facto} model for image generation, yet the heavy training overhead hinders its broader adoption in the research community. We observe that diffusion models are commonly trained to learn all fine-grained visual information from scratch. This paradigm may cause unnecessary training costs hence requiring in-depth investigation. In this work, we show that it suffices to train a strong diffusion model by first pre-training the model to learn some primer distribution that loosely characterizes the unknown real image distribution. Then the pre-trained model can be fine-tuned for various generation tasks efficiently. In the pre-training stage, we propose to mask a high proportion (e.g., up to 90\%) of input images to approximately represent the primer distribution and introduce a masked denoising score matching objective to train a model to denoise visible areas. In subsequent fine-tuning stage, we efficiently train diffusion model without masking. Utilizing the two-stage training framework, we achieves significant training acceleration and a new FID score record of 6.27 on CelebA-HQ $256 \times 256$ for ViT-based diffusion models. The generalizability of a pre-trained model further helps building models that perform better than ones trained from scratch on different downstream datasets. For instance, a diffusion model pre-trained on VGGFace2 attains a 46\% quality improvement when fine-tuned on a different dataset that contains only 3000 images. Our code is available at \url{https://github.com/jiachenlei/maskdm}.

cs.CV↗

arXiv2023

Flat F-theory and friends

We discuss F-theory backgrounds associated to flat torus bundles over Ricci-flat manifolds. In this setting the F-theory background can be understood as a IIB orientifold with a large radius limit described by a supersymmetric compactification of IIB supergravity on a smooth, Ricci flat, but in general non-spin geometry. When compactified on an additional circle these backgrounds are T-dual to IIA compactifications on smooth non-orientable manifolds with a Pin-minus structure.

hep-th↗