arXiv subjects

Wenyuan Liu

Publications and source records attributed to Wenyuan Liu.

At least 19 recordsLinked to original sources

arXiv2026

ARCQuant: Boosting NVFP4 Quantization with Augmented Residual Channels for LLMs

The emergence of fine-grained numerical formats like NVFP4 presents new opportunities for efficient Large Language Model (LLM) inference. However, it is difficult to adapt existing Post-Training Quantization (PTQ) strategies to these formats: rotation-based methods compromise fine-grained block isolation; smoothing techniques struggle with significant 4-bit quantization errors; and mixed-precision approaches often conflict with hardware constraints on unified-precision computation. To address these challenges, we propose ARCQuant, a framework that boosts NVFP4 performance via Augmented Residual Channels. Distinct from methods that compromise block isolation or hardware uniformity, ARCQuant maintains a strictly unified NVFP4 format by augmenting the activation matrix with quantized residual channels. This design integrates the error compensation process directly into the matrix reduction dimension, enabling the use of standard, highly optimized GEMM kernels with minimal overhead. Theoretical analysis confirms that the worst-case error bound of our dual-stage NVFP4 quantization is comparable to that of standard 8-bit formats such as MXFP8. Extensive experiments on LLaMA and Qwen models demonstrate that ARCQuant achieves state-of-the-art accuracy, comparable to full-precision baselines in perplexity and downstream tasks. Furthermore, deployment on RTX 5090 and RTX PRO 6000 GPUs confirms practical benefits, achieving up to 3x speedup over FP16. Our code is available at https://github.com/actypedef/ARCQuant.

cs.LG↗

arXiv2026

SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference

Low-bit floating-point formats and semi-structured sparsity are increasingly supported by modern accelerators, yet combining them for LLM activation compression remains challenging: activations contain input-dependent outliers that dominate block scales in FP4 quantization, and directly applying N:M sparsity masks discards moderate values, coupling sparsification loss with quantization error. We introduce SharQ, a training-free inference method that bridges activation sparsity and FP4 quantization through an online sparse--dense decomposition. For each activation tensor, SharQ generates an input-adaptive N:M mask to extract an outlier-dominated sparse backbone, quantizes it to FP4, and defines a dense residual relative to the quantized sparse backbone rather than the unquantized sparse values. A sparse FP4 GEMM processes the backbone while a dense FP4 GEMM compensates for both mask-induced activation loss and sparse-path quantization error. The two paths share a single FP4 weight payload with path-specific scale views, and a fused preparation kernel absorbs mask generation, residual construction, and layer normalization into one operator. SharQ requires no calibration data, retraining, or model-specific tuning. Evaluated on Llama-3.1-8B, Qwen2.5-7B, Qwen3-30B-A3B, and Qwen3-VL-8B, SharQ recovers 43--63% of the NVFP4-to-FP16 accuracy gap across language and vision-language tasks, and generalizes across NVFP4, HiF4, and MXFP4 formats. On an RTX 5090, SharQ delivers 2.2--2.4$\times$ latency reduction over FP16 and 1.2--1.4$\times$ throughput improvement over FP8 in language model serving, and up to 1.58$\times$ speedup on Wan2.2-T2V-A14B video generation when combined with SageAttention. Our code is available at https://github.com/actypedef/SharQ.

cs.LG↗

arXiv2026

MicroMix: Efficient Mixed-Precision Quantization with Microscaling Formats for Large Language Models

Quantization significantly accelerates inference in large language models (LLMs) by replacing original high-precision matrices with low-precision counterparts. Recent advances in weight-activation quantization have primarily focused on mapping both weights and activations to the INT4 format. Although the new FP4 Tensor Cores in NVIDIA's Blackwell architecture offer up to 4x speedup over FP16, existing INT4-based kernels fail to fully exploit this capability due to mismatched data formats. To bridge this gap, we propose MicroMix, a co-designed mixed-precision quantization algorithm and GEMM kernel based on Microscaling (MX) data formats. Tailored for the Blackwell architecture, the MicroMix kernel supports arbitrary combinations of MXFP4, MXFP6, and MXFP8 channels, and produces BFloat16 outputs. To achieve a favorable trade-off between accuracy and efficiency for each linear layer, we introduce quantization thresholds that identify activation elements where lower-precision formats (MXFP4 or MXFP6) incur excessive quantization error. Our algorithm selectively allocates higher-precision channels to preserve accuracy while maintaining compute efficiency. On the Llama and Qwen model families, MicroMix achieves near-FP16 performance across diverse downstream tasks with an average precision of 5 bits. In particular, Qwen2.5-32B-Base, Coder and Math exhibit lossless accuracy on zero-shot, code generation, and mathematical reasoning benchmarks. In addition, on RTX 5070Ti laptop and RTX 5090 GPUs, our kernel achieves 2.29-3.38x acceleration compared to TensorRT-FP16. Our code is available at https://github.com/lwy2020/MicroMix.

cs.LG↗

arXiv2026

Post-Training Quantization of OpenPangu Models for Efficient Deployment on Atlas A2

Huawei's openPangu-Embedded-1B and openPangu-Embedded-7B are variants of the openPangu large language model, designed for efficient deployment on Ascend NPUs. The 7B variant supports three distinct Chain-of-Thought (CoT) reasoning paradigms, namely slow_think, auto_think, and no_think, while the 1B variant operates exclusively in the no_think mode, which employs condensed reasoning for higher efficiency. Although CoT reasoning enhances capability, the generation of extended reasoning traces introduces substantial memory and latency overheads, posing challenges for practical deployment on Ascend NPUs. This paper addresses these computational constraints by leveraging low-bit quantization, which transforms FP16 computations into more efficient integer arithmetic. We introduce a unified low-bit inference framework, supporting INT8 (W8A8) and W4A8 quantization, specifically optimized for openPangu-Embedded models on the Atlas A2. Our comprehensive evaluation on code generation benchmarks (HumanEval and MBPP) demonstrates the efficacy of this approach. INT8 quantization consistently preserves over 90\% of the FP16 baseline accuracy and achieves a 1.5x prefill speedup on the Atlas A2. Furthermore, W4A8 quantization significantly reduces memory consumption, albeit with a moderate trade-off in accuracy. These findings collectively indicate that low-bit quantization effectively facilitates efficient CoT reasoning on Ascend NPUs, maintaining high model fidelity.

cs.LG↗

arXiv2025

Routing Distilled Knowledge via Mixture of LoRA Experts for Large Language Model based Bundle Generation

Large Language Models (LLMs) have shown potential in automatic bundle generation but suffer from prohibitive computational costs. Although knowledge distillation offers a pathway to more efficient student models, our preliminary study reveals that naively integrating diverse types of distilled knowledge from teacher LLMs into student LLMs leads to knowledge conflict, negatively impacting the performance of bundle generation. To address this, we propose RouteDK, a framework for routing distilled knowledge through a mixture of LoRA expert architecture. Specifically, we first distill knowledge from the teacher LLM for bundle generation in two complementary types: high-level knowledge (generalizable rules) and fine-grained knowledge (session-specific reasoning). We then train knowledge-specific LoRA experts for each type of knowledge together with a base LoRA expert. For effective integration, we propose a dynamic fusion module, featuring an input-aware router, where the router balances expert contributions by dynamically determining optimal weights based on input, thereby effectively mitigating knowledge conflicts. To further improve inference reliability, we design an inference-time enhancement module to reduce variance and mitigate suboptimal reasoning. Experiments on three public datasets show that our RouteDK achieves accuracy comparable to or even better than the teacher LLM, while maintaining strong computational efficiency. In addition, it outperforms state-of-the-art approaches for bundle generation.

cs.CL↗

arXiv2025

Does Knowledge Distillation Matter for Large Language Model based Bundle Generation?

LLMs are increasingly explored for bundle generation, thanks to their reasoning capabilities and knowledge. However, deploying large-scale LLMs introduces significant efficiency challenges, primarily high computational costs during fine-tuning and inference due to their massive parameterization. Knowledge distillation (KD) offers a promising solution, transferring expertise from large teacher models to compact student models. This study systematically investigates knowledge distillation approaches for bundle generation, aiming to minimize computational demands while preserving performance. We explore three critical research questions: (1) how does the format of KD impact bundle generation performance? (2) to what extent does the quantity of distilled knowledge influence performance? and (3) how do different ways of utilizing the distilled knowledge affect performance? We propose a comprehensive KD framework that (i) progressively extracts knowledge (patterns, rules, deep thoughts); (ii) captures varying quantities of distilled knowledge through different strategies; and (iii) exploits complementary LLM adaptation techniques (in-context learning, supervised fine-tuning, combination) to leverage distilled knowledge in small student models for domain-specific adaptation and enhanced efficiency. Extensive experiments provide valuable insights into how knowledge format, quantity, and utilization methodologies collectively shape LLM-based bundle generation performance, exhibiting KD's significant potential for more efficient yet effective LLM-based bundle generation.

cs.CL↗

arXiv2025

Adaptive In-Context Learning with Large Language Models for Bundle Generation

Most existing bundle generation approaches fall short in generating fixed-size bundles. Furthermore, they often neglect the underlying user intents reflected by the bundles in the generation process, resulting in less intelligible bundles. This paper addresses these limitations through the exploration of two interrelated tasks, i.e., personalized bundle generation and the underlying intent inference, based on different user sessions. Inspired by the reasoning capabilities of large language models (LLMs), we propose an adaptive in-context learning paradigm, which allows LLMs to draw tailored lessons from related sessions as demonstrations, enhancing the performance on target sessions. Specifically, we first employ retrieval augmented generation to identify nearest neighbor sessions, and then carefully design prompts to guide LLMs in executing both tasks on these neighbor sessions. To tackle reliability and hallucination challenges, we further introduce (1) a self-correction strategy promoting mutual improvements of the two tasks without supervision signals and (2) an auto-feedback mechanism for adaptive supervision based on the distinct mistakes made by LLMs on different neighbor sessions. Thereby, the target session can gain customized lessons for improved performance by observing the demonstrations of its neighbor sessions. Experiments on three real-world datasets demonstrate the effectiveness of our proposed method.

cs.IR↗

arXiv2024

CrossQuant: A Post-Training Quantization Method with Smaller Quantization Kernel for Precise Large Language Model Compression

Post-Training Quantization (PTQ) is an effective technique for compressing Large Language Models (LLMs). While many studies focus on quantizing both weights and activations, it is still a challenge to maintain the accuracy of LLM after activating quantization. To investigate the primary cause, we extend the concept of kernel from linear algebra to quantization functions to define a new term, "quantization kernel", which refers to the set of elements in activations that are quantized to zero. Through quantitative analysis of the quantization kernel, we find that these elements are crucial for maintaining the accuracy of quantized LLMs. With the decrease of quantization kernel, the precision of quantized LLMs increases. If the quantization kernel proportion is kept below 19% for OPT models and below 1% for LLaMA models, the precision loss from quantizing activations to INT8 becomes negligible. Motivated by the goal of developing a quantization method with small quantization kernel, we propose CrossQuant: a simple yet effective method for quantizing activations. CrossQuant cross-quantizes elements using row and column-wise absolute maximum vectors, achieving a quantization kernel of approximately 16% for OPT models and less than 0.1% for LLaMA models. Experimental results on LLMs (LLaMA, OPT) ranging from 6.7B to 70B parameters demonstrate that CrossQuant improves or maintains perplexity and accuracy in language modeling, zero-shot, and few-shot tasks.

cs.LG↗

arXiv2024

3D-RPE: Enhancing Long-Context Modeling Through 3D Rotary Position Encoding

Inspired by the Bloch Sphere representation, we propose a novel rotary position encoding on a three-dimensional sphere, named 3D Rotary Position Encoding (3D-RPE). 3D-RPE is an advanced version of the widely used 2D Rotary Position Encoding (RoPE), with two major advantages for modeling long contexts: controllable long-term decay and improved position resolution. For controllable long-term decay, 3D-RPE allows for the regulation of long-term decay within the chunk size, ensuring the modeling of relative positional information between tokens at a distant relative position. For enhanced position resolution, 3D-RPE can mitigate the degradation of position resolution caused by position interpolation on RoPE. We have conducted experiments on long-context Natural Language Understanding (NLU) and long-sequence Language Modeling (LM) tasks. From the experimental results, 3D-RPE achieved performance improvements over RoPE, especially in long-context NLU tasks.

cs.CL↗

arXiv2022

Is there evidence for exponential quantum advantage in quantum chemistry?

The idea to use quantum mechanical devices to simulate other quantum systems is commonly ascribed to Feynman. Since the original suggestion, concrete proposals have appeared for simulating molecular and materials chemistry through quantum computation, as a potential ``killer application''. Indications of potential exponential quantum advantage in artificial tasks have increased interest in this application, thus, it is critical to understand the basis for potential exponential quantum advantage in quantum chemistry. Here we gather the evidence for this case in the most common task in quantum chemistry, namely, ground-state energy estimation. We conclude that evidence for such an exponential advantage across chemical space has yet to be found. While quantum computers may still prove useful for quantum chemistry, it may be prudent to assume exponential speedups are not generically available for this problem.

physics.chem-ph↗

arXiv2022

Strategic differences between regional investments into graphene technology and how corporations and universities manage patent portfolios

Nowadays, patenting activities are essential in converting applied science to technology in the prevailing innovation model. To gain strategic advantages in the technological competitions between regions, nations need to leverage the investments of public and private funds to diversify over all technologies or specialize in a small number of technologies. In this paper, we investigated who the leaders are at the regional and assignee levels, how they attained their leadership positions, and whether they adopted diversification or specialization strategies, using a dataset of 176,193 patent records on graphene between 1986 and 2017 downloaded from Derwent Innovation. By applying a co-clustering method to the IPC subclasses in the patents and using a z-score method to extract keywords from their titles and abstracts, we identified seven graphene technology areas emerging in the sequence synthesis - composites - sensors - devices - catalyst - batteries - water treatment. We then examined the top regions in their investment preferences and their changes in rankings over time and found that they invested in all seven technology areas. In contrast, at the assignee level, some were diversified while others were specialized. We found that large entities diversified their portfolios across multiple technology areas, while small entities specialized around their core competencies. In addition, we found that universities had higher entropy values than corporations on average, leading us to the hypothesis that corporations file, buy, or sell patents to enable product development. In contrast, universities focus only on licensing their patents. We validated this hypothesis through an aggregate analysis of reassignment and licensing and a more detailed analysis of three case studies - SAMSUNG, RICE UNIVERSITY, and DYSON.

cs.DL↗

arXiv2021

Silicon photonics integrated dynamic polarization controller for continuous-variable quantum key distribution

We designed and demonstrated experimentally a silicon photonics integrated dynamic polarization controller which is a crucial component of a continuous-variable quantum key distribution system. By using a variable step simulated annealing approach, we achieve a dynamic polarization extinction ratio greater than 25 dB. The dynamic polarization controller can be utilized in silicon photonics integrated continuous-variable quantum key distribution system to minimize the size and decrease the cost further.

quant-ph↗

arXiv2021

Coevolution of theoretical and applied research: a case study of graphene research by temporal and geographic analysis

As a part of science of science (SciSci) research, the evolution of scientific disciplines has been attracting a great deal of attention recently. This kind of discipline level analysis not only give insights of one particular field but also shed light on general principles of scientific enterprise. In this paper we focus on graphene research, a fast growing field covers both theoretical and applied study. Using co-clustering method, we split graphene literature into two groups and confirm that one group is about theoretical research (T) and another corresponds to applied research (A). We analyze the proportion of T/A and found applied research becomes more and more popular after 2007. Geographical analysis demonstrated that countries have different preference in terms of T/A and they reacted differently to research trend. The interaction between two groups has been analyzed and shows that T extremely relies on T and A heavily relies on A, however the situation is very stable for T but changed markedly for A. No geographic difference is found for the interaction dynamics. Our results give a comprehensive picture of graphene research evolution and also provide a general framework which is able to analyze other disciplines.

cs.DL↗

arXiv2020

Spatial-Temporal Dataset of COVID-19 Outbreak in China

We present Coronavirus disease 2019 (COVID-19) statistics in China dataset: daily statistics of the COVID-19 outbreak in China at the city/county level. For each city/country, we include the six most important numbers for epidemic research: daily new infections, accumulated infections, daily new recoveries, accumulated recoveries, daily new deaths, and accumulated deaths. We cross validate the dataset and the estimate error rate is about 0.04%. We then give several examples to show how to trace the spreading in particular cities or provinces, and also contrast the development of COVID-19 in all cities in China at the early, middle and late stages. We hope this dataset can help researchers around the world better understand the spreading dynamics of COVID-19 at a regional level, to inform intervention and mitigation strategies for policymakers.

q-bio.PE↗

arXiv2019

The role of mainstreamness and interdisciplinarity for the relevance of scientific papers

There is demand from science funders, industry, and the public that science should become more risk-taking, more out-of-the-box, and more interdisciplinary. Is it possible to tell how interdisciplinary and out-of-the-box scientific papers are, or which papers are mainstream? Here we use the bibliographic coupling network, derived from all physics papers that were published in the Physical Review journals in the past century, to try to identify them as mainstream, out-of-the-box, or interdisciplinary. We show that the network clusters into scientific fields. The position of individual papers with respect to these clusters allows us to estimate their degree of mainstreamness or interdisciplinary. We show that over the past decades the fraction of mainstream papers increases, the fraction of out-of-the-box decreases, and the fraction of interdisciplinary papers remains constant. Studying the rewards of papers, we find that in terms of absolute citations, both, mainstream and interdisciplinary papers are rewarded. In the long run, mainstream papers perform less than interdisciplinary ones in terms of citation rates. We conclude that to avoid a trend towards mainstreamness a new incentive scheme is necessary.

cs.DL↗

arXiv2019

The Golden Eras of Graphene Science and Technology: Bibliographic Evidences From Journal and Patent Publications

Today's scientific research is an expensive enterprise funded largely by taxpayers' and corporate groups' monies. It is a critical part in the competition between nations, and all nations want to discover fields of research that promise to create future industries, and dominate these by building up scientific and technological expertise early. However, our understanding of the value chain going from science to technology is still in a relatively infant stage, and the conversion of scientific leadership into market dominance remains very much an alchemy rather than a science. In this paper, we analyze bibliometric records of scientific journal publications and patents related to graphene, at the aggregate level as well as on the temporal and spatial dimensions. We find the present leaders of graphene science and technology emerged rather late in the race, after the initial scientific leaders lost their footings. More importantly, notwithstanding the amount of funding already committed, we find evidences that suggest the 'Golden Eras' of graphene science and technology were in 2010 and 2012 respectively, in spite of the continued growth of journal and patent publications in this area.

cs.DL↗

arXiv2018

Using Machine Learning to Predict the Evolution of Physics Research

The advancement of science as outlined by Popper and Kuhn is largely qualitative, but with bibliometric data it is possible and desirable to develop a quantitative picture of scientific progress. Furthermore it is also important to allocate finite resources to research topics that have growth potential, to accelerate the process from scientific breakthroughs to technological innovations. In this paper, we address this problem of quantitative knowledge evolution by analysing the APS publication data set from 1981 to 2010. We build the bibliographic coupling and co-citation networks, use the Louvain method to detect topical clusters (TCs) in each year, measure the similarity of TCs in consecutive years, and visualize the results as alluvial diagrams. Having the predictive features describing a given TC and its known evolution in the next year, we can train a machine learning model to predict future changes of TCs, i.e., their continuing, dissolving, merging and splitting. We found the number of papers from certain journals, the degree, closeness, and betweenness to be the most predictive features. Additionally, betweenness increases significantly for merging events, and decreases significantly for splitting events. Our results represent a first step from a descriptive understanding of the Science of Science (SciSci), towards one that is ultimately prescriptive.

physics.soc-ph↗

arXiv2018

PEPS++: Towards Extreme-scale Simulations of Strongly Correlated Quantum Many-particle Models on Sunway TainhuLight

The study of strongly frustrated magnetic systems has drawn great attentions from both theoretical and experimental physics. Efficient simulations of these models are essential for understanding their exotic properties. Here we present PEPS++, a novel computational paradigm for simulating frustrated magnetic systems and other strongly correlated quantum many-body systems. PEPS++ can accurately solve these models at the extreme scale with low cost and high scalability on modern heterogeneous supercomputers. We implement PEPS++ on Sunway TaihuLight based on a carefully designed tensor computation library for manipulating high-rank tensors and optimize it by invoking various high-performance matrix and tensor operations. By solving a 2D strongly frustrated $J_1$-$J_2$ model with over ten million cores, PEPS++ demonstrates the capability of simulating strongly correlated quantum many-body problems at unprecedented scales with accuracy and time-to-solution far beyond the previous state of the art.

cond-mat.str-el↗