arXiv subjects

Qi Huang

Publications and source records attributed to Qi Huang.

At least 55 records · Page 3Linked to original sources

arXiv2025

MegaScale-MoE: Large-Scale Communication-Efficient Training of Mixture-of-Experts Models in Production

We present MegaScale-MoE, a production system tailored for the efficient training of large-scale mixture-of-experts (MoE) models. MoE emerges as a promising architecture to scale large language models (LLMs) to unprecedented sizes, thereby enhancing model performance. However, existing MoE training systems experience a degradation in training efficiency, exacerbated by the escalating scale of MoE models and the continuous evolution of hardware. Recognizing the pivotal role of efficient communication in enhancing MoE training, MegaScale-MoE customizes communication-efficient parallelism strategies for attention and FFNs in each MoE layer and adopts a holistic approach to overlap communication with computation at both inter- and intra-operator levels. Additionally, MegaScale-MoE applies communication compression with adjusted communication patterns to lower precision, further improving training efficiency. When training a 352B MoE model on 1,440 NVIDIA Hopper GPUs, MegaScale-MoE achieves a training throughput of 1.41M tokens/s, improving the efficiency by 1.88$\times$ compared to Megatron-LM. We share our operational experience in accelerating MoE training and hope that by offering our insights in system design, this work will motivate future research in MoE systems.

cs.LG↗

arXiv2025

The recoil corrections, correlation functions and possible double-strange hadronic molecules

In this work, we perform a systematic investigation of ${K}^{(*)}{K}^{(*)}$ interactions within a one-boson-exchange model.The framework incorporates both $S-D$ wave mixing and coupled-channel effects, with effective potentials retained up to order $\mathcal{O}(1/M^2)$. By solving the coupled channel Schrödinger equations, we can predict two double-strange molecular candidates: a $KK^*$ molecule with $I(J^P)=0(1^+)$ and a $K^*K^*$ molecule with $0(1^+)$. Our results also show that the recoil corrections play a crucial role in the formation of these two molecular candidates. Furthermore, the $S-D$ wave mixing effects contribute positively to the formation process. As a byproduct of this analysis, we extend our study to $K^{(*)}\bar{K}^{(*)}$ interactions with the same model. Our findings suggest that the $K\bar{K}^{*}$ states with $0(1^{+-}, 1^{++})$ and $K^{*}\bar{K}^{*}$ states with $0(0^{++}, 1^{+-}, 2^{++})$ can be promising molecular candidates. Additionally, we analyze the correlations between the constituent mesons, the resulting correlation functions provide additional support for our predictions of both double-strange and strangonium-like molecular states.

hep-ph↗

arXiv2025

Roles of $\bar{D}^{}K^{}$ and $D^\bar{D}$ molecular states in decay $B^+ \to D^{+} D^- K^+$

This study investigates the three-body decay process $B^+ \to D^{*+} D^- K^+$, aiming to explore the possible origins of $T^*_{\bar{c}\bar{s}0}(2870)^0$ and $χ_{c1}(3872)$ as intermediate states. Within the molecular state framework, $T^*_{\bar{c}\bar{s}0}(2870)^0$ and $χ_{c1}(3872)$ are considered as possible $\bar{D}^{*}K^{}$ and $D^*\bar{D}$ molecular states, respectively. Using effective Lagrangians, the interaction kernels of the $\bar{D}^{*}K^{*}$ and $D^*\bar{D}$ systems are constructed within the one-boson-exchange model. The corresponding rescattering amplitudes and pole positions are obtained by solving the quasipotential Bethe-Salpeter equation. These amplitudes are incorporated into the decay amplitude of the three-body process, and the $D^-K^+$ and $D^{*+}D^-$ invariant mass spectra are simulated via Monte Carlo methods. To better reproduce the experimental data, additional Breit-Wigner contributions from $T^*_{\bar{c}\bar{s}1}(2900)^0$, $χ_{c1}(4010)$, and $h_c(4300)$ are included. The results show a pronounced enhancement near 2900 MeV in the $D^-K^+$ invariant mass spectrum, strongly supporting the interpretation of $T^*_{\bar{c}\bar{s}0}(2870)^0$ as a $\bar{D}^{*}K^{*}$ molecular state. While the $\bar{D}^{*}K^{*}$ molecular state provides a reasonable contribution to the $D^-K^+$ spectrum, the $D^*\bar{D}$ molecular state yields no significant effect on either the $D^-K^+$ or $D^{*+}D^-$ distributions. This suggests that the observed $χ_{c1}(3872)$ structure around 3872 MeV may not be interpreted as a $D^*\bar{D}$ molecular state.

hep-ph↗

arXiv2025

Emergence of charm-strange dibaryons with negative parity via baryon-baryon interactions

Within the framework of the one-boson-exchange model, we systematically perform a coupled channel analysis of the $P-$wave interactions between a charm baryon and light baryon, the involved channels include $Ξ_cN$, $Λ_cΣ$, $Ξ_c^{\prime}N$, $Σ_cΛ$, $Ξ_c^*N$, $Σ_c^*Λ$, $Σ_cΣ$, and $Σ_c^*Σ$. Our results can predict several possible molecular candidates, such as a $Ξ_c^{\prime}N$ molecule with $I(J^P)=0(1^-)$, $Ξ_c^{*}N$ molecules with $0(0^-, 1^-, 2^-)$, $Σ_cΣ$ molecules with $0(1^-)$ and $1(1^-)$, and $Σ_c^*Σ$ molecules with $0(0^-, 1^-, 2^-)$, and $1(2^-)$. The coupled channel effects significantly influence the formation of the $Σ_cΣ$ molecule with $1(1^-)$. Furthermore, we analyze the phase shifts for these coupled channel systems. Our analysis not only confirms the existence of the predicted molecules but also identifies potential resonant dibaryons, including a $Σ_cΣ$ shape-type resonance with $1(1^-)$, a $Σ_c^*Σ$ shape-type resonance with $1(0^-)$, a $Λ_cΣ/Σ_cΣ$ coupled Feshbach-type resonance with $1(1^-)$, and $Λ_cΣ/Σ_c^*Σ$ coupled Feshbach-type resonances with $1(0^-, 2^-)$.

hep-ph↗

arXiv2025

Superradiance Enhanced Light-Matter Interaction in Spatially Ordered Shape and Volume Controlled Single Quantum Dots: Enabling On-Chip Photonic Networks

On-chip photonic networks require adequately spatially ordered matter-photon interconversion qubit sources with emission figures-of-merit exceeding the requirements that would enable the desired functional response of the network. The mesa-top single quantum dots (MTSQDs) have recently been demonstrated to meet these requirements. The substrate-encoded size-reducing epitaxy (SESRE) approach underpinning the realization of these unique quantum emitters allows control on the shape, size, and strain (lattice-matched or mismatched) of these epitaxial single quantum dots. We have exploited this unique feature of the MTSQDs to reproducibly create arrays of quantum dots that exhibit single photon superradiance, a characteristic of the SESRE-enabled delicate balance between the confinement pontential volume, depth, the resulting exciton binding energy, and the degree of confinement of the center of mass (CM) motion of the exciton. Scanning transmission electron microscope (STEM) studies reveal the structural (atomic scale) and chemical (nm scale) nature of the material region defining the notion of the shape and volume (here large) of the electron confinement region (i.e. the QD). In the exciton's weake CM confinement regime, owing to its coherent sampling of the large volume, enhancement of the MTSQD oscillator strength to ~30 is demonstrated. Theoretical modelling with input from the STEM finding provide corroboration for single photon superradiance causing enhancement of the oscillator strength by ~2.5 to 3. Our findings allow fabricating and studying interconnected networks enabled by the unique matter qubit-light qubit interconversion units that can be realized for lattice matched and mismatched material combinations covering UV to midinfrared wavelength range.

cond-mat.mes-hall↗

arXiv2025

Intent Classification on Low-Resource Languages with Query Similarity Search

Intent classification is an important component of a functional Information Retrieval ecosystem. Many current approaches to intent classification, typically framed as a classification problem, can be problematic as intents are often hard to define and thus data can be difficult and expensive to annotate. The problem is exacerbated when we need to extend the intent classification system to support multiple and in particular low-resource languages. To address this, we propose casting intent classification as a query similarity search problem - we use previous example queries to define an intent, and a query similarity method to classify an incoming query based on the labels of its most similar queries in latent space. With the proposed approach, we are able to achieve reasonable intent classification performance for queries in low-resource languages in a zero-shot setting.

cs.IR↗

arXiv2025

Why Are You Wrong? Counterfactual Explanations for Language Grounding with 3D Objects

Combining natural language and geometric shapes is an emerging research area with multiple applications in robotics and language-assisted design. A crucial task in this domain is object referent identification, which involves selecting a 3D object given a textual description of the target. Variability in language descriptions and spatial relationships of 3D objects makes this a complex task, increasing the need to better understand the behavior of neural network models in this domain. However, limited research has been conducted in this area. Specifically, when a model makes an incorrect prediction despite being provided with a seemingly correct object description, practitioners are left wondering: "Why is the model wrong?". In this work, we present a method answering this question by generating counterfactual examples. Our method takes a misclassified sample, which includes two objects and a text description, and generates an alternative yet similar formulation that would have resulted in a correct prediction by the model. We have evaluated our approach with data from the ShapeTalk dataset along with three distinct models. Our counterfactual examples maintain the structure of the original description, are semantically similar and meaningful. They reveal weaknesses in the description, model bias and enhance the understanding of the models behavior. Theses insights help practitioners to better interact with systems as well as engineers to improve models.

cs.AI↗

arXiv2025

Prediction of $p\barΩ$ states and femtoscopic study

Inspired by recent researches on the $p Ω$ and $p \barΛ$ systems, we investigate the $p \barΩ$ systems within the framework of a quark model. Our results show that the attraction between a nucleon and $\barΩ$ is slightly stronger than that between a nucleon and $Ω$, suggesting that the $p \barΩ$ system is more likely to form bound states. The dynamic calculations indicate that the $p \barΩ$ systems with both $J^{P}=1^{-}$ and $2^{-}$ can form bound states, with binding energies deeper than those of the $p Ω$ systems with $J^{P}=2^{+}$. The scattering phase shift and scattering parameter calculations also support the existence of $p \barΩ$ states. Additionally, we discuss the behavior of the femtoscopic correlation function for the $p \barΩ$ pairs for the first time. Considering the significant progress in experimental measurements of the correlation function of the $p Ω$ system, the further study of the $p \barΩ$ systems using femtoscopic techniques will be a very valuable work.

hep-ph↗

arXiv2025

Seed1.5-Thinking: Advancing Superb Reasoning Models with Reinforcement Learning

We introduce Seed1.5-Thinking, capable of reasoning through thinking before responding, resulting in improved performance on a wide range of benchmarks. Seed1.5-Thinking achieves 86.7 on AIME 2024, 55.0 on Codeforces and 77.3 on GPQA, demonstrating excellent reasoning abilities in STEM and coding. Beyond reasoning tasks, the method demonstrates notable generalization across diverse domains. For instance, it surpasses DeepSeek R1 by 8% in win rate on non-reasoning tasks, indicating its broader applicability. Compared to other state-of-the-art reasoning models, Seed1.5-Thinking is a Mixture-of-Experts (MoE) model with a relatively small size, featuring 20B activated and 200B total parameters. As part of our effort to assess generalized reasoning, we develop two internal benchmarks, BeyondAIME and Codeforces, both of which will be publicly released to support future research. Model trial link: https://www.volcengine.com/experience/ark.

cs.CL↗

arXiv2025

Exploring the nature of $η_{1}(1855)$ and it's partner in a chiral quark model

Inspired by the recent experimental discoveries of $X(3872)$ ($c\bar{q}$-$q\bar{c}$) and $T_{cc}$ ($c\bar{q}$-$c\bar{q}$), we systematically study two four-quark systems: the $K K_1$ ($q\bar{s}$-$q\bar{s}$) system and the $K \bar{K}_1$ ($q\bar{s}$-$s\bar{q}$) system, which is a candidate for the recently observed $η_1(1855)$. Within the framework of an accurate few-body calculation method (GEM), we employ the chiral quark model to simultaneously consider the molecular and diquark structures of these two multiquark systems and include their channel coupling effects. Our results show that the $K \bar{K}_1$ system remains a scattering state. On the other hand, due to the presence of a good-diquark structure in the $K K_1$ system, we obtain a bound state in the coupled-channel calculation. The primary contribution to the binding energy comes from the exchange of $π$-meson and $σ$-meson. The inter-quark distance indicates that it is a compact four-quark structure.

hep-ph↗

arXiv2025

Strong decays of singly heavy baryons

More and more excited baryons have been reported experimentally, but many properties are still unclear. This work attempts to simultaneously study the masses and strong decay widths of some singly heavy baryons, in order to provide possible quantum numbers for these states. The chiral quark model and the $^{3}P_{0}$ decay model are employed to calculate the masses and decay widths of $Λ_{c(b)}$ and $Σ_{c(b)}$ baryons for all quantum numbers with $2S$, $1P$, and $2P$ waves. We considered not only two-body strong decays but also the influence of three-body decays. Our calculations show that: (i) For states with experimentally determined quantum numbers, such as $Λ_c(2595)$, $Λ_c(2625)$, $Λ_b(5912)$ and $Λ_b(5920)$, the results are consistent with experimental data and the conclusions of most theoretical studies. (ii) For states whose quantum numbers have not yet been fully determined experimentally, we provide possible interpretations. For example, our calculations tend to interpret $Λ_c(2910)$ is interpreted as a $J^P=\frac{3}{2}^-$ state with 1P-wave $ρ$-mode or a $J^P=\frac{1}{2}^-$ state with 2P wave $λ$-mode. $Λ_c(2940)$ can be interpreted as the $J^P=\frac{3}{2}^-$ state with 2P-wave $λ$-mode. For $Λ_c(2860)$, we offer a different interpretation, proposing that its mass and width closely match those of a 2P-wave $J^P=\frac{1}{2}^{-}$ state. It is hoped that our calculations can provide valuable information for the experimental and theoretical studies of heavy baryons.

hep-ph↗

arXiv2025

Spectral Analysis and Decay Mechanisms of $1^{-+}$ Hybrid States in Light Meson Sector

The exploration of exotic mesons, which transcend the conventional quark-antiquark framework, is pivotal for advancing our understanding of QCD and the strong interaction. Among these, states possessing the quantum numbers $J^{PC}=1^{-+}$, such as $π_1(1600)$, $π_1(2015)$, and the recently discovered $η_1(1855)$, have attracted significant attention due to their potential hybrid nature, which involves gluonic excitations. However, the physical interpretation of these states remains contentious, primarily due to inconsistencies between theoretical predictions and experimental observations regarding their decay widths and mass spectra. To address these challenges, we employ a potential model inspired by SU$(3)$ lattice gauge theory to calculate the masses of light-flavor $1^{-+}$ hybrid states and utilize the constituent gluon model to analyze their strong decay properties. Our mass calculations suggest that while the $π_1(1600)$ and $η_1(1855)$ masses are in close agreement with the predicted hybrid states, the corresponding decay widths for $π_1(1600)$ and $π_1(2015)$ do not support their classification as hybrids, implying alternative interpretations, such as tetraquark or molecular configurations. Conversely, the $η_1(1855)$ is consistent with a mixed hybrid state composed of $(u\bar{u}+d\bar{d})g/\sqrt{2}$ and $s\bar{s}g$ components, with mixing angles ranging from $17.7^\circ$ to $84.2^\circ$. Additionally, we predict the masses and decay widths of yet-unobserved isoscalar and excited hybrid states, providing valuable targets for future experimental searches. This study not only clarifies the hybrid nature of certain exotic mesons but also contributes to the systematic classification of hybrid states within the meson nonet, thereby enhancing our comprehension of exotic hadronic states.

hep-ph↗

arXiv2025

Investigating the nature of $N(1535)$ and $Λ(1405)$ in a quenched chiral quark model

In this work, we systematically study $N(1440)$, $N(1535)$, and $Λ(1405)$ in both the quenched three-quark and five-quark frameworks using the Gaussian Expansion Method (GEM) within the chiral quark model. Our calculations show that $N(1535)$ can be reproduced as a three-quark state ($N(1P)$), while $N(1440)$ and $Λ(1405)$ cannot be accommodated as the three-quark candidates, ($N(2S)$ and $Λ(1P)$), respectively. In the five-quark framework, we find that the $ΛK$ state for $N(1535)$ can not form a bound state, while in the $N\bar{K}$ channel there will $Λ(1405)$ form a shallow bound state. Based on the complex-scaling method, we performed complete coupled-channels calculations and obtained six resonance states with energies ranging from 1.8 GeV to 2.2 GeV, in addition with one bound state located around $Σπ$ channel. However, neither molecular candidates in $ΛK$ channel for $N(1535)$ nor $N\bar{K}$ for $Λ(1405)$ are included in these states. This is because the strong coupling between $N\bar{K}$ and $Σπ$ will make the $N\bar{K}$ unbound, while the weak coupling between $ΛK$ and $ΣK$ can not help form a stable structure around $ΛK$ threshold. Thus, under the quenched quark model, our results support $N(1535)$ as a three-quark state, while $N(1440)$ is neither a three-quark nor a five-quark state. In addition, we find that although $Λ(1405)$ can be primarily a five-quark state, it requires a mixture of three-quark and five-quark components for stability. In the future, an exploration on the mixing effects between bare baryons with these relevant two-body hadronic channel components will be carried out to further test our conclusions.

hep-ph↗

arXiv2025

P2DFlow: A Protein Ensemble Generative Model with SE(3) Flow Matching

Biological processes, functions, and properties are intricately linked to the ensemble of protein conformations, rather than being solely determined by a single stable conformation. In this study, we have developed P2DFlow, a generative model based on SE(3) flow matching, to predict the structural ensembles of proteins. We specifically designed a valuable prior for the flow process and enhanced the model's ability to distinguish each intermediate state by incorporating an additional dimension to describe the ensemble data, which can reflect the physical laws governing the distribution of ensembles, so that the prior knowledge can effectively guide the generation process. When trained and evaluated on the MD datasets of ATLAS, P2DFlow outperforms other baseline models on extensive experiments, successfully capturing the observable dynamic fluctuations as evidenced in crystal structure and MD simulations. As a potential proxy agent for protein molecular simulation, the high-quality ensembles generated by P2DFlow could significantly aid in understanding protein functions across various scenarios. Code is available at https://github.com/BLEACH366/P2DFlow

physics.bio-ph↗

arXiv2025

ByteScale: Efficient Scaling of LLM Training with a 2048K Context Length on More Than 12,000 GPUs

Scaling long-context ability is essential for Large Language Models (LLMs). To amortize the memory consumption across multiple devices in long-context training, inter-data partitioning (a.k.a. Data Parallelism) and intra-data partitioning (a.k.a. Context Parallelism) are commonly used. Current training frameworks predominantly treat the two techniques as orthogonal, and establish static communication groups to organize the devices as a static mesh (e.g., a 2D mesh). However, the sequences for LLM training typically vary in lengths, no matter for texts, multi-modalities or reinforcement learning. The mismatch between data heterogeneity and static mesh causes redundant communication and imbalanced computation, degrading the training efficiency. In this work, we introduce ByteScale, an efficient, flexible, and scalable LLM training framework for large-scale mixed training of long and short sequences. The core of ByteScale is a novel parallelism strategy, namely Hybrid Data Parallelism (HDP), which unifies the inter- and intra-data partitioning with a dynamic mesh design. In particular, we build a communication optimizer, which eliminates the redundant communication for short sequences by data-aware sharding and dynamic communication, and further compresses the communication cost for long sequences by selective offloading. Besides, we also develop a balance scheduler to mitigate the imbalanced computation by parallelism-aware data assignment. We evaluate ByteScale with the model sizes ranging from 7B to 141B, context lengths from 256K to 2048K, on a production cluster with more than 12,000 GPUs. Experiment results show that ByteScale outperforms the state-of-the-art training system by up to 7.89x.

cs.DC↗

arXiv2025

Beyond the Veil of Similarity: Quantifying Semantic Continuity in Explainable AI

We introduce a novel metric for measuring semantic continuity in Explainable AI methods and machine learning models. We posit that for models to be truly interpretable and trustworthy, similar inputs should yield similar explanations, reflecting a consistent semantic understanding. By leveraging XAI techniques, we assess semantic continuity in the task of image recognition. We conduct experiments to observe how incremental changes in input affect the explanations provided by different XAI methods. Through this approach, we aim to evaluate the models' capability to generalize and abstract semantic concepts accurately and to evaluate different XAI methods in correctly capturing the model behaviour. This paper contributes to the broader discourse on AI interpretability by proposing a quantitative measure for semantic continuity for XAI methods, offering insights into the models' and explainers' internal reasoning processes, and promoting more reliable and transparent AI systems.

cs.AI↗

arXiv2025

Wafer-scale Integration of Single-Crystalline MoS$_2$ for Flexible Electronics Enabled by Oxide Dry-transfer

Atomically thin, single-crystalline transition metal dichalcogenides (TMDCs) grown via chemical vapor deposition (CVD) on sapphire substrates exhibit exceptional mechanical and electrical properties, positioning them as excellent channel materials for flexible electronics. However, conventional wet-transfer processes for integrating these materials onto flexible substrates often introduce surface contamination, significantly degrading device performance. Here, we present a wafer-scale dry-transfer technique using a high-dielectric oxide as the transfer medium, enabling the integration of 4-inch single-crystalline MoS$_2$ onto flexible substrates. This method eliminates contact with polymers or solvents, thus preserving the intrinsic electronic properties of MoS$_2$. As a result, the fabricated flexible field-effect transistor (FET) arrays exhibit remarkable performance, with a mobility of 117 cm$^2$/Vs, a subthreshold swing of 68.8 mV dec$^{-1}$, and an ultra-high current on/off ratio of $10^{12}$-values comparable to those achieved on rigid substrates. Leveraging the outstanding electrical characteristics, we demonstrated MoS$_2$-based flexible inverters operating in the subthreshold regime, achieving both a high gain of 218 and ultra-low power consumption of 1.4 pW/$μ$m. Additionally, we integrated a flexible tactile sensing system driven by active-matrix MoS$_2$ FET arrays onto a robotic gripper, enabling real-time object identification. These findings demonstrate the simultaneous achievement of high electrical performance and flexibility, highlighting the immense potential of single-crystalline TMDC-based flexible electronics for real-world applications.

physics.app-ph↗

arXiv2025

Investigating the $p$-$Ω$ Interaction and Correlation Functions

Motivated by experimental measurements, we investigate the $p$-$Ω$ correlation functions and interactions on the basis of a quark model. By solving the inverse scattering problem with channel coupling, we renormalize the coupling to other channels into an effective single-channel $p$-$Ω$ potentials. The effects of Coulomb interaction and spin-averaging are also discussed. According to our results, the depletion of the $p$-$Ω$ correlation functions, which is attributed to the $J^P = 2^+$ bound state not observed in the ALICE Collaboration's measurements [Nature \textbf{588}, 232 (2020)], can be explained by the contribution of the attractive $J^P = 1^+$ component in spin-averaging. So far, we have provided a consistent description of the $p$-$Ω$ system from the perspective of the quark model, including the energy spectrum, scattering phase shifts, and correlation functions. The existence of the $p$-$Ω$ bound state has been supported by all three aspects. Additionally, a sign of the $p$-$Ω$ correlation function's subtle sub-unity part can be seen in experimental measurements, which warrants more precise verification in the future.

hep-ph↗