Search

Large Language Models (LLMs) have achieved remarkable success across numerous domains. However, the high time complexity of existing pruning and quantization methods significantly hinders their effective deployment on resource-constrained consumer or edge devices. In this study, we propose a novel Hessian-free Weight Pruning-Quantization (HWPQ) method. HWPQ eliminates the need for computationally intensive Hessian matrix calculations by introducing a contribution-based weight metric, which evaluates the importance of weights without relying on second-order derivatives. Additionally, we employ the Exponentially Weighted Moving Average (EWMA) technique to bypass weight sorting, enabling the selection of weights that contribute most to LLM accuracy and further reducing time complexity. Our approach is extended to support 2:4 structured sparsity pruning, facilitating efficient execution on modern hardware accelerators. Experimental results demonstrate that HWPQ significantly enhances the compression performance of LLaMA2. Compared to state-of-the-art quantization and pruning frameworks, HWPQ achieves average speedups of 5.97x (up to 20.75x) in quantization time and 12.29x (up to 56.02x) in pruning time, while largely preserving model accuracy. Furthermore, we observe a 1.50x inference speedup compared to the baseline.

Since its discovery in 1977, the spin-parity of $\Xi(2030)$ has not been fully determined experimentally. The latest Particle Data Group (PDG) listing suggests it may be a baryon with $J=5/2$. Therefore, studying the mass spectrum and decay properties of $\Xi(2030)$ has become a current hot topic to definitively establish its spin-parity. As the three-quark model fails to explain $\Xi(2030)$, we previously proposed it may be a molecule primarily composed of $\bar{K}^{}\Sigma$ with $J^P=5/2^{+}$, based on its mass spectrum study. To verify its molecular state interpretation, this work proposes studying the strong decays of $\Xi(2030)$ assuming it is a $P$-wave $J^P=5/2^{+}$ meson-baryon molecule predominantly composed of $\bar{K}^{}\Sigma$. We calculated all experimentally measured two-body and three-body final state decay widths of $\Xi(2030)$, including $\Xi(2030) \to \bar{K}\Lambda, \bar{K}\Sigma, \pi\Xi, \pi\Xi^{*}$, and $\Xi(2030) \to \pi\pi\Xi, \pi\bar{K}\Sigma, \pi\bar{K}\Lambda$. The results indicate that both the total decay width and partial decay widths agree well with experimental values within the error margins. This supports that $\Xi(2030)$ is a molecule with spin-parity $J^P = 5/2^{+}$, predominantly composed of $\bar{K}^{*}\Sigma$. Compared to the experimental central values, our results are slightly smaller, which suggests that $\Xi(2030)$ may contain additional components besides meson-baryon molecular components, such as three quark structures., Comment: 8 papers,6 figures,1 table

Heavy quark symmetry can help us identify the internal structure of hadrons and predict new particles. In this study, we examine the strong decay modes of the observed $\Lambda_b^0(6146)$ and $\Lambda_b^0(6152)$, assuming these two states are molecular states primarily composed of $\bar{B}^{*}N$ component. The partial decay widths of the $\bar{B}^{*}N$ molecular state into the $\pi\Sigma_b$ and $\pi\Sigma_b^{*}$ final states through hadronic loops are calculated using effective Lagrangians. Our results, when compared with LHCb observations, support the interpretation of $\Lambda_b^0(6146)$ as a molecule primarily composed of $\bar{B}^{*}N$ components. However, the decay width of $\Lambda_b^0(6152)$ cannot be accurately reproduced within the molecular state framework. Based on the above results and heavy quark symmetry, we predict the existence of $\bar{B}^{*}N$ molecular states with $J^p=5/2^{+}$, which are the heavy quark spin symmetry partners of $\Lambda_b(6146)$, with masses in the range of 6195-6200 MeV. And the main decay is $\pi\Sigma_b^{*}$ channel. Moreover, there must existence of a $D^{*}N$ molecule with $J^p=3/2^{+}$, possible corresponding to the experimentally observed $\Lambda_c(2860)^{+}$. If $\Lambda_c(2880)^{+}$ is indeed the heavy quark flavor symmetry partner of $\Lambda_b(6152)$, it would exhibit a conventional three-quark structure. Therefore, we also propose the search for a $D^{*}N$ molecule with a spin-parity of $J^p=5/2^{+}$, which would be the heavy-quark spin partner state of $\Lambda_c(2860)^{+}$. It should be noted that these baryons may be mixed states, containing both molecular and three-quark components. These results can aid experiments in exploring the internal structure of these baryons.

Video anomaly understanding (VAU) aims to automatically comprehend unusual occurrences in videos, thereby enabling various applications such as traffic surveillance and industrial manufacturing. While existing VAU benchmarks primarily concentrate on anomaly detection and localization, our focus is on more practicality, prompting us to raise the following crucial questions: "what anomaly occurred?", "why did it happen?", and "how severe is this abnormal event?". In pursuit of these answers, we present a comprehensive benchmark for Causation Understanding of Video Anomaly (CUVA). Specifically, each instance of the proposed benchmark involves three sets of human annotations to indicate the "what", "why" and "how" of an anomaly, including 1) anomaly type, start and end times, and event descriptions, 2) natural language explanations for the cause of an anomaly, and 3) free text reflecting the effect of the abnormality. In addition, we also introduce MMEval, a novel evaluation metric designed to better align with human preferences for CUVA, facilitating the measurement of existing LLMs in comprehending the underlying cause and corresponding effect of video anomalies. Finally, we propose a novel prompt-based method that can serve as a baseline approach for the challenging CUVA. We conduct extensive experiments to show the superiority of our evaluation metric and the prompt-based approach. Our code and dataset are available at https://github.com/fesvhtr/CUVA., Comment: Accepted in CVPR2024, Codebase: https://github.com/fesvhtr/CUVA

We show that Seiberg-like duality of $\mathcal{N}=1$ gauge theory coupled with tensor chiral fields and fundamental chiral fields works if the meson spectrum built from the tensor fields takes particular form: a) It should be truncated; b) The $R$ charges of tensor fields $\{R_a\}$ and the truncated mesons $\{R_j\}$ take very special values. The meson spectrum so that the duality works is encoded elegantly in the factorization of the polynomial $y^n-1=\Phi_{+}\Phi_{-}$. Our consideration covers many known $\mathcal{N}=1$ dualities and generates a large class of new examples., Comment: 31 pages, 6 figures

The discovery of gravitational waves (GWs) opens a new window for exploring the physics of the early universe. Identifying the source of GWs and their spectra today turn out to be the important tasks so as to assist the experimental detection of stochastic GWs. In this paper, we investigate the oscillations of the ultralight dark photon (ULDP) into GWs in the dark halo. Assuming dark matter is composed of the ULDP and there are primordial dark magnetic fields (PDMFs) arising from the axion inflation and/or the dark phase transition, then the ULDP can oscillate into the GW when it passes through an environment of PDMFs. We derive the local energy density of GWs in the galaxy cluster induced by the instaneous oscillation of ULDP in the PDMFs. These stochastic local GWs exhibit a pulse-like spectrum, with frequency depending on the mass of the ULDP, and can be detected in Pulsar Timing Arrays (PTAs) or future space-based interferometers. We also find that the low-frequency GW signal observed by the NANOGrav collaboration and other PTA experiments can be explained by the oscillation of the ULDP in the PDMFs in the early universe., Comment: 7 pages, 3 figures

Axion-like particles (ALPs) are promising dark matter candidates. Their signals in direct detection experiments arise from the well-known inverse Primakoff effect or the inverse Compton scattering of the ALPs with the electron. In this paper, we revisit the direct detection of ALP by carefully considering the interference between the inverse Primakoff amplitude and the inverse Compton amplitude in the scattering process $a+e \to e+\gamma$ for the first time. It shows that the contribution of the interference term turns to be dominated in the scattering for a large ALP energy. Given the new analytical formula, signals or constraints of ALP couplings in various projected experiments are investigated. Our results show that these experiments may put strong constraints on ALP couplings for relatively heavy ALP. We further study projected constraints on the ALP from the JUNO experiment, which shows competitive constraints on ALP couplings using a ten-year exposure.

We systematically study the AdS/CFT correspondence induced by D3 branes probing three dimensional Gorenstein quotient singularity $\mathbb{C}^3/G$. The field theory is given by the McKay quiver, which has a vanishing NSVZ beta function assuming that all the chiral fields have the $U(1)_R$ charge $\frac{2}{3}$. Various physical quantities such as quiver Hilbert series, superconformal index, central charges, etc are computed, which match exactly with those computed using the singularity. We also study the relevant deformation of those theories and find the dual geometry, therefore generate many new interesting AdS/CFT pairs. The quiver gauge theory defined using finite subgroups of $SO(3)$ group has some interesting features, for example, its Seiberg duality behavior is quite interesting., Comment: 43 pages, 15 tables, and 13 figures

Recently, a new doubly charged tetraquark $T^{a}_{c\bar{s}0}(2900)^{++}$ and its neutral partner $T^{a}_{c\bar{s}0}(2900)^0$ at the invariant mass spectrum of $\pi{}D_s$ were observed by the LHCb Collaboration. According to its properties, such as the mass and decay width, the $T^{a}_{c\bar{s}0}(2900)^{++/0}$ have been suggested to be a compact multi-quark state or a hadron molecule. In order to distinguish the various interpretations of the $T^{a}_{c\bar{s}0}(2900)^{++/0}$, we investigate the possibility to study the $\bar{T}^{a}_{c\bar{s}0}(2900)$ [the antiparticle of $T^{a}_{c\bar{s}0}(2900)$] by kaon-induced reactions on a proton target in an effective Lagrangian approach. The production mechanism is characterized by the $t$-channel $D$ meson exchange. Our theoretical approach is based on the assumption that $\bar{T}^{a}_{c\bar{s}0}(2900)$ can be either a $K^{*}D^{*}-D_s^{*}\rho$ molecule or a compact tetraquark state. Using the coupling constants of the $\bar{T}^{a}_{c\bar{s}0}(2900)$ to $KD$ channel obtained from molecule or compact tetraquark picture of the $\bar{T}^{a}_{c\bar{s}0}(2900)$, we compute the cross-sections for the process $K^{-}n\to{}\bar{T}^{a}_{c\bar{s}0}(2900)^{--}\Lambda^{+}_c$. The $\bar{K}N$ initial state interaction mediated by Pomeron and Reggeon exchanges is also included, which reduces the production of the $\bar{T}^{a}_{c\bar{s}0}(2900)$. Our calculations show that whether $\bar{T}^{a}_{c\bar{s}0}(2900)$ is a molecule or a compact tetraquark state, the cross-sections for the $K^{-}n\to{}\bar{T}^{a}_{c\bar{s}0}(2900)^{--}\Lambda^{+}_c$ reaction are of similar magnitude, ranging from approximately 0.075 nb to 0.270 nb.