arXiv subjects

Ting Cao

Publications and source records attributed to Ting Cao.

At least 109 records · Page 6Linked to original sources

arXiv2024

Proximity-Induced Exchange Interaction: a New Pathway for Quantum Sensing using Spin Centers in Hexagonal Boron Nitride

Defects in hexagonal boron nitride (hBN), a two-dimensional van der Waals material, have raised wide range interest for its potential in various quantum applications. Due to hBN's 2D nature, spin center in hBN can be engineered in close proximity to target material, providing advantages over their 3D counterparts, such as nitrogen-vacancy (NV) center in diamond. Here we propose a novel quantum sensing protocol driven by exchange interaction between spin center in hBN and the underlying magnetic substrate induced by magnetic proximity effect. By first-principle calculation, we demonstrate the induced exchange interaction dominates over dipole-dipole interaction by orders of magnitude when in proximity. The interaction remains antiferromagnetic across all stacking configuration between the spin center in hBN and the target van der Waals magnets. Additionally, we explored the scaling behavior of the exchange field as a function of the spatial separation between the spin center and the targets.

cond-mat.mes-hall↗

arXiv2024

Polarization-driven band topology evolution in twisted MoTe$_2$ and WSe$_2$

Motivated by recent experimental observations of opposite Chern numbers in $R$-type twisted MoTe$_2$ and WSe$_2$ homobilayers, we perform large-scale density-functional-theory (DFT) calculations with machine learning force fields to investigate moiré band topology from large to small twist angles in both materials. We find that the Chern numbers of the moiré frontier bands change sign as a function of twist angle, and this change is driven by the competition between moiré ferroelectricity and piezoelectricity. Our large-scale calculations, enabled by machine learning methods, reveal crucial insights into interactions across different scales in twisted bilayer systems. The interplay between atomic-level relaxation effects and moiré-scale electrostatic potential variation opens new avenues for the design of intertwined topological and correlated states, including the possibility of mimicking higher Landau-level physics in the absence of a magnetic field.

cond-mat.mtrl-sci↗

arXiv2024

BitDistiller: Unleashing the Potential of Sub-4-Bit LLMs via Self-Distillation

The upscaling of Large Language Models (LLMs) has yielded impressive advances in natural language processing, yet it also poses significant deployment challenges. Weight quantization has emerged as a widely embraced solution to reduce memory and computational demands. This paper introduces BitDistiller, a framework that synergizes Quantization-Aware Training (QAT) with Knowledge Distillation (KD) to boost the performance of LLMs at ultra-low precisions (sub-4-bit). Specifically, BitDistiller first incorporates a tailored asymmetric quantization and clipping technique to maximally preserve the fidelity of quantized weights, and then proposes a novel Confidence-Aware Kullback-Leibler Divergence (CAKLD) objective, which is employed in a self-distillation manner to enable faster convergence and superior model performance. Empirical evaluations demonstrate that BitDistiller significantly surpasses existing methods in both 3-bit and 2-bit configurations on general language understanding and complex reasoning benchmarks. Notably, BitDistiller is shown to be more cost-effective, demanding fewer data and training resources. The code is available at https://github.com/DD-DuDa/BitDistiller.

cs.CL↗

arXiv2023

Fractional Chern Insulator in Twisted Bilayer MoTe$_2$

A recent experiment has reported the first observation of a zero-field fractional Chern insulator (FCI) phase in twisted bilayer MoTe$_2$ moiré superlattices [Nature 622, 63-68 (2023)]. The experimental observation is at an unexpected large twist angle 3.7$^\circ$ and calls for a better understanding of the FCI in real materials. In this work, we perform large-scale density functional theory calculation for the twisted bilayer MoTe$_2$, and find that lattice reconstruction is crucial for the appearance of an isolated flat Chern band. The existence of the FCI state at $ν= -2/3$ are confirmed by exact diagonalization. We establish phase diagrams with respect to the twist angle and electron interaction, which reveal an optimal twist angle of $3.5^\circ$ for the observation of FCI. We further demonstrate that an external electric field can destroy the FCI state by changing band geometry and show evidence of the $ν=-3/5$ FCI state in this system. Our research highlights the importance of accurate single particle band structure in the quest for strong correlated electronic states and provides insights into engineering fractional Chern insulator in moiré superlattices.

cond-mat.str-el↗

arXiv2023

AFPQ: Asymmetric Floating Point Quantization for LLMs

Large language models (LLMs) show great performance in various tasks, but face deployment challenges from limited memory capacity and bandwidth. Low-bit weight quantization can save memory and accelerate inference. Although floating-point (FP) formats show good performance in LLM quantization, they tend to perform poorly with small group sizes or sub-4 bits. We find the reason is that the absence of asymmetry in previous FP quantization makes it unsuitable for handling asymmetric value distribution of LLM weight tensors. In this work, we propose asymmetric FP quantization (AFPQ), which sets separate scales for positive and negative values. Our method leads to large accuracy improvements and can be easily plugged into other quantization methods, including GPTQ and AWQ, for better performance. Besides, no additional storage is needed compared with asymmetric integer (INT) quantization. The code is available at https://github.com/zhangsichengsjtu/AFPQ.

cs.CL↗

arXiv2023

Raman Study of Layered Breathing Kagome Lattice Semiconductor Nb3Cl8

Niobium chloride (Nb3Cl8) is a layered 2D semiconducting material with many exotic properties including a breathing kagome lattice, a topological flat band in its band structure, and a crystal structure that undergoes a structural and magnetic phase transition at temperatures below 90 K. Despite being a remarkable material with fascinating new physics, the understanding of its phonon properties is at its infancy. In this study, we investigate the phonon dynamics of Nb3Cl8 in bulk and few layer flakes using polarized Raman spectroscopy and density functional theory (DFT) analysis to determine the material's vibrational modes, as well as their symmetrical representations and atomic displacements. We experimentally resolved 12 phonon modes, 5 of which are A1g modes while the remaining 7 are Eg modes, which is in strong agreement with our DFT calculation. Layer-dependent results suggest that the Raman peak positions are mostly insensitive to changes in layer thickness, while peak intensity and FWHM are affected. Raman measurements as a function of excitation wavelength (473-785 nm) show a significant increase of the peak intensities when using a 473 nm excitation source, suggesting a near resonant condition. Temperature-dependent Raman experiments carried out above and below the transition temperature did not show any change in the symmetries of the phonon modes, suggesting that the structural phase transition is likely from the high temperature P3m1 phase to the low-temperature R3m phase. Magneto-Raman measurements carried out at 140 and 2 K between -2 to 2 T show that the Raman modes are not magnetically coupled. Overall, our study presented here significantly advances the fundamental understanding of layered Nb3Cl8 material which can be further exploited for future applications.

cond-mat.mtrl-sci↗

arXiv2023

Observation of flat and weakly dispersing bands in a van der Waals semiconductor Nb3Br8 with breathing kagome lattice

Niobium halides, Nb3X8 (X = Cl,Br,I), which are predicted two-dimensional magnets, have recently gotten attention due to their breathing kagome geometry. Here, we have studied the electronic structure of Nb3Br8 by using angle-resolved photoemission spectroscopy (ARPES) and first-principles calculations. ARPES results depict the presence of multiple flat and weakly dispersing bands. These bands are well explained by the theoretical calculations, which show they have Nb d character indicating their origination from the Nb atoms forming the breathing kagome plane. This van der Waals material can be easily thinned down via mechanical exfoliation to the ultrathin limit and such ultrathin samples are stable as depicted from the time-dependent Raman spectroscopy measurements at room temperature. These results demonstrate that Nb3Br8 is an excellent material not only for studying breathing kagome induced flat band physics and its connection with magnetism, but also for heterostructure fabrication for application purposes.

cond-mat.mes-hall↗

arXiv2023

LUT-NN: Empower Efficient Neural Network Inference with Centroid Learning and Table Lookup

On-device Deep Neural Network (DNN) inference consumes significant computing resources and development efforts. To alleviate that, we propose LUT-NN, the first system to empower inference by table lookup, to reduce inference cost. LUT-NN learns the typical features for each operator, named centroid, and precompute the results for these centroids to save in lookup tables. During inference, the results of the closest centroids with the inputs can be read directly from the table, as the approximated outputs without computations. LUT-NN integrates two major novel techniques: (1) differentiable centroid learning through backpropagation, which adapts three levels of approximation to minimize the accuracy impact by centroids; (2) table lookup inference execution, which comprehensively considers different levels of parallelism, memory access reduction, and dedicated hardware units for optimal performance. LUT-NN is evaluated on multiple real tasks, covering image and speech recognition, and nature language processing. Compared to related work, LUT-NN improves accuracy by 66% to 92%, achieving similar level with the original models. LUT-NN reduces the cost at all dimensions, including FLOPs ($\leq$ 16x), model size ($\leq$ 7x), latency ($\leq$ 6.8x), memory ($\leq$ 6.5x), and power ($\leq$ 41.7%).

cs.LG↗

arXiv2023

Boosting DNN Cold Inference on Edge Devices

DNNs are ubiquitous on edge devices nowadays. With its increasing importance and use cases, it's not likely to pack all DNNs into device memory and expect that each inference has been warmed up. Therefore, cold inference, the process to read, initialize, and execute a DNN model, is becoming commonplace and its performance is urgently demanded to be optimized. To this end, we present NNV12, the first on-device inference engine that optimizes for cold inference NNV12 is built atop 3 novel optimization knobs: selecting a proper kernel (implementation) for each DNN operator, bypassing the weights transformation process by caching the post-transformed weights on disk, and pipelined execution of many kernels on asymmetric processors. To tackle with the huge search space, NNV12 employs a heuristic-based scheme to obtain a near-optimal kernel scheduling plan. We fully implement a prototype of NNV12 and evaluate its performance across extensive experiments. It shows that NNV12 achieves up to 15.2x and 401.5x compared to the state-of-the-art DNN engines on edge CPUs and GPUs, respectively.

cs.LG↗

arXiv2023

Gate-tunable antiferromagnetic Chern insulator in twisted bilayer transition metal dichalcogenides

A series of recent experimental works on twisted MoTe$_2$ homobilayers have unveiled an abundance of exotic states in this system. Valley-polarized quantum anomalous Hall states have been identified at hole doping of $ν= -1$, and the fractional quantum anomalous Hall effect is observed at $ν= -2/3$ and $ν= -3/5$. In this work, we investigate the electronic properties of AA-stacked twisted bilayer MoTe$_2$ at $ν=-2$ by $k$-space Hartree-Fock calculations. We find that the phase diagram is qualitatively similar to the phase diagram of a Kane-Mele-Hubbard with staggered onsite potential. A noteworthy phase within the diagram is the antiferromagnetic Chern insulator, stabilized by the external electric field. We attribute the existence of this Chern insulator to an antiferromagnetic instability at a topological phase transition between the quantum spin hall phase and a band insulator phase. We highlight that the antiferromagnetic Chern insulator phase is most evident at a twist angle of approximately $4^\circ$. Our research proposes the potential of realizing a Chern insulator beyond $ν=-1$, and contributes fresh perspectives on the interplay between band topology and electron-electron correlations in moiré superlattices.

cond-mat.mes-hall↗

arXiv2023

Observation of Fractionally Quantized Anomalous Hall Effect

The integer quantum anomalous Hall (QAH) effect is a lattice analog of the quantum Hall effect at zero magnetic field. This striking transport phenomenon occurs in electronic systems with topologically nontrivial bands and spontaneous time-reversal symmetry breaking. Discovery of its putative fractional counterpart in the presence of strong electron correlations, i.e., the fractional quantum anomalous Hall (FQAH) effect, would open a new chapter in condensed matter physics. Here, we report the direct observation of both integer and fractional QAH effects in electrical measurements on twisted bilayer MoTe$_2$. At zero magnetic field, near filling factor $ν= -1$ (one hole per moiré unit cell) we see an extended integer QAH plateau in the Hall resistance $R_\text{xy}$ that is quantized to $h/e^2 \pm 0.1 \%$ while the longitudinal resistance $R_\text{xx}$ vanishes. Remarkably, at $ν=-2/3$ and $-3/5$ we see plateau features in $R_\text{xy}$ at $3h/2e^2 \pm 1\%$ and $5h/3e^2 \pm 3\%$, respectively, while $R_\text{xx}$ remains small. All these features shift linearly in an applied magnetic field with slopes matching the corresponding Chern numbers $-1$, $-2/3$, and $-3/5$, precisely as expected for integer and fractional QAH states. In addition, at zero magnetic field, $R_\text{xy}$ is approximately $2h/e^2$ near half filling ($ν= -1/2$) and varies linearly as $ν$ is tuned. This behavior resembles that of the composite Fermi liquid in the half-filled lowest Landau level of a two-dimensional electron gas at high magnetic field. Direct observation of the FQAH and associated effects paves the way for researching charge fractionalization and anyonic statistics at zero magnetic field.

cond-mat.mes-hall↗

arXiv2023

A Quantized Interband Topological Index in Two-Dimensional Systems

We introduce a novel gauge-invariant, quantized interband index in two-dimensional (2D) multiband systems. It provides a bulk topological classification of a submanifold of parameter space (e.g., an electron valley in a Brillouin zone), and therefore overcomes difficulties in characterizing topology of submanifolds. We confirm its topological nature by numerically demonstrating a one-to-one correspondence to the valley Chern number in $k\cdot p$ models (e.g., gapped Dirac fermion model), and the first Chern number in lattice models (e.g., Haldane model). Furthermore, we derive a band-resolved topological charge and demonstrate that it can be used to investigate the nature of edge states due to band inversion in valley systems like multilayer graphene.

cond-mat.mes-hall↗

arXiv2023

Constraint-aware and Ranking-distilled Token Pruning for Efficient Transformer Inference

Deploying pre-trained transformer models like BERT on downstream tasks in resource-constrained scenarios is challenging due to their high inference cost, which grows rapidly with input sequence length. In this work, we propose a constraint-aware and ranking-distilled token pruning method ToP, which selectively removes unnecessary tokens as input sequence passes through layers, allowing the model to improve online inference speed while preserving accuracy. ToP overcomes the limitation of inaccurate token importance ranking in the conventional self-attention mechanism through a ranking-distilled token distillation technique, which distills effective token rankings from the final layer of unpruned models to early layers of pruned models. Then, ToP introduces a coarse-to-fine pruning approach that automatically selects the optimal subset of transformer layers and optimizes token pruning decisions within these layers through improved $L_0$ regularization. Extensive experiments on GLUE benchmark and SQuAD tasks demonstrate that ToP outperforms state-of-the-art token pruning and model compression methods with improved accuracy and speedups. ToP reduces the average FLOPs of BERT by 8.1x while achieving competitive accuracy on GLUE, and provides a real latency speedup of up to 7.4x on an Intel CPU.

cs.CL↗

arXiv2023

Accurate and Structured Pruning for Efficient Automatic Speech Recognition

Automatic Speech Recognition (ASR) has seen remarkable advancements with deep neural networks, such as Transformer and Conformer. However, these models typically have large model sizes and high inference costs, posing a challenge to deploy on resource-limited devices. In this paper, we propose a novel compression strategy that leverages structured pruning and knowledge distillation to reduce the model size and inference cost of the Conformer model while preserving high recognition performance. Our approach utilizes a set of binary masks to indicate whether to retain or prune each Conformer module, and employs L0 regularization to learn the optimal mask values. To further enhance pruning performance, we use a layerwise distillation strategy to transfer knowledge from unpruned to pruned models. Our method outperforms all pruning baselines on the widely used LibriSpeech benchmark, achieving a 50% reduction in model size and a 28% reduction in inference cost with minimal performance loss.

cs.CL↗

arXiv2023

Adam Accumulation to Reduce Memory Footprints of both Activations and Gradients for Large-scale DNN Training

Running out of GPU memory has become a main bottleneck for large-scale DNN training. How to reduce the memory footprint during training has received intensive research attention. We find that previous gradient accumulation reduces activation memory but fails to be compatible with gradient memory reduction due to a contradiction between preserving gradients and releasing gradients. To address this issue, we propose a novel optimizer accumulation method for Adam, named Adam Accumulation (AdamA), which enables reducing both activation and gradient memory. Specifically, AdamA directly integrates gradients into optimizer states and accumulates optimizer states over micro-batches, so that gradients can be released immediately after use. We mathematically and experimentally demonstrate AdamA yields the same convergence properties as Adam. Evaluated on transformer-based models, AdamA achieves up to 23% memory reduction compared to gradient accumulation with less than 2% degradation in training throughput. Notably, AdamA can work together with memory reduction methods for optimizer states to fit 1.26x~3.14x larger models over PyTorch and DeepSpeed baseline on GPUs with different memory capacities.

cs.LG↗

arXiv2023

Integer or Floating Point? New Outlooks for Low-Bit Quantization on Large Language Models

Efficient deployment of large language models (LLMs) necessitates low-bit quantization to minimize model size and inference cost. While low-bit integer formats (e.g., INT8/INT4) have been the conventional choice, emerging low-bit floating-point formats (e.g., FP8/FP4) offer a compelling alternative and are gaining support from cutting-edge hardware, such as NVIDIA's H100 GPU. However, the superiority of low-bit INT versus FP formats for quantization on LLMs remains unclear. In this study, we conduct a comparative analysis of INT and FP quantization with the same bit-width, revealing that the optimal quantization format varies across different layers due to the complexity and diversity of tensor distribution. Consequently, we advocate the Mixture of Formats Quantization (MoFQ), which selects the optimal format on a layer-wise basis. This simple yet effective approach achieves state-of-the-art results in both weight-only (W-only) and weight-activation (WA) post-training quantization scenarios when tested on LLaMA across various tasks. In 4-bit W-only quantization, MoFQ surpasses GPTQ without complex hyperparameter tuning and with an order of magnitude faster quantization speed. While in 8-bit WA quantization, MoFQ significantly outperforms INT/FP-only methods, achieving performance close to the full precision model. Notably, MoFQ incurs no hardware overhead compared to INT/FP-only quantization, as the bit-width remains unchanged.

cs.LG↗

arXiv2023

Gate-tunable phonon magnetic moment in bilayer graphene

We develop a first-principles quantum scheme to calculate the phonon magnetic moment in solids. As a showcase example, we apply our method to study gated bilayer graphene, a material with strong covalent bonds. According to the classical theory based on the Born effective charge, the phonon magnetic moment in this system should vanish, yet our quantum mechanical calculations find significant phonon magnetic moments. Furthermore, the magnetic moment is highly tunable by changing the gate voltage. Our results firmly establish the necessity of the quantum mechanical treatment, and identify small-gap covalent materials as a promising platform for studying tunable phonon magnetic moment.

cond-mat.mtrl-sci↗

arXiv2023

Signatures of Fractional Quantum Anomalous Hall States in Twisted MoTe2 Bilayer

The interplay between spontaneous symmetry breaking and topology can result in exotic quantum states of matter. A celebrated example is the quantum anomalous Hall (QAH) state, which exhibits an integer quantum Hall effect at zero magnetic field thanks to its intrinsic ferromagnetism. In the presence of strong electron-electron interactions, exotic fractional-QAH (FQAH) states at zero magnetic field can emerge. These states could host fractional excitations, including non-Abelian anyons - crucial building blocks for topological quantum computation. Flat Chern bands are widely considered as a desirable venue to realize the FQAH state. For this purpose, twisted transition metal dichalcogenide homobilayers in rhombohedral stacking have recently been predicted to be a promising material platform. Here, we report experimental signatures of FQAH states in 3.7-degree twisted MoTe2 bilayer. Magnetic circular dichroism measurements reveal robust ferromagnetic states at fractionally hole filled moiré minibands. Using trion photoluminescence as a sensor, we obtain a Landau fan diagram which shows linear shifts in carrier densities corresponding to the v=-2/3 and -3/5 ferromagnetic states with applied magnetic field. These shifts match the Streda formula dispersion of FQAH states with fractionally quantized Hall conductance of -2/3$e^2/h$ and -3/5$e^2/h$, respectively. Moreover, the v=-1 state exhibits a dispersion corresponding to Chern number -1, consistent with the predicted QAH state. In comparison, several non-ferromagnetic states on the electron doping side do not disperse, i.e., are trivial correlated insulators. The observed topological states can be further electrically driven into topologically trivial states. Our findings provide clear evidence of the long-sought FQAH states, putting forward MoTe2 moiré superlattices as a fascinating platform for exploring fractional excitations.

cond-mat.mes-hall↗