arXiv subjects

Shu Liu

Publications and source records attributed to Shu Liu.

At least 73 records · Page 4Linked to original sources

arXiv2025

Hunyuan3D 2.1: From Images to High-Fidelity 3D Assets with Production-Ready PBR Material

3D AI-generated content (AIGC) is a passionate field that has significantly accelerated the creation of 3D models in gaming, film, and design. Despite the development of several groundbreaking models that have revolutionized 3D generation, the field remains largely accessible only to researchers, developers, and designers due to the complexities involved in collecting, processing, and training 3D models. To address these challenges, we introduce Hunyuan3D 2.1 as a case study in this tutorial. This tutorial offers a comprehensive, step-by-step guide on processing 3D data, training a 3D generative model, and evaluating its performance using Hunyuan3D 2.1, an advanced system for producing high-resolution, textured 3D assets. The system comprises two core components: the Hunyuan3D-DiT for shape generation and the Hunyuan3D-Paint for texture synthesis. We will explore the entire workflow, including data preparation, model architecture, training strategies, evaluation metrics, and deployment. By the conclusion of this tutorial, you will have the knowledge to finetune or develop a robust 3D generative model suitable for applications in gaming, virtual reality, and industrial design.

cs.CV↗

arXiv2025

ARPO:End-to-End Policy Optimization for GUI Agents with Experience Replay

Training large language models (LLMs) as interactive agents for controlling graphical user interfaces (GUIs) presents a unique challenge to optimize long-horizon action sequences with multimodal feedback from complex environments. While recent works have advanced multi-turn reinforcement learning (RL) for reasoning and tool-using capabilities in LLMs, their application to GUI-based agents remains relatively underexplored due to the difficulty of sparse rewards, delayed feedback, and high rollout costs. In this paper, we investigate end-to-end policy optimization for vision-language-based GUI agents with the aim of improving performance on complex, long-horizon computer tasks. We propose Agentic Replay Policy Optimization (ARPO), an end-to-end RL approach that augments Group Relative Policy Optimization (GRPO) with a replay buffer to reuse the successful experience across training iterations. To further stabilize the training process, we propose a task selection strategy that filters tasks based on baseline agent performance, allowing the agent to focus on learning from informative interactions. Additionally, we compare ARPO with offline preference optimization approaches, highlighting the advantages of policy-based methods in GUI environments. Experiments on the OSWorld benchmark demonstrate that ARPO achieves competitive results, establishing a new performance baseline for LLM-based GUI agents trained via reinforcement learning. Our findings underscore the effectiveness of reinforcement learning for training multi-turn, vision-language GUI agents capable of managing complex real-world UI interactions. Codes and models:https://github.com/dvlab-research/ARPO.git.

cs.CV↗

arXiv2025

DCO$^+$ and DCN 1-0 survey toward a sample of Planck cold clumps

Deuterated molecules can be used to study the physical conditions and the astro-chemical evolution of molecular clouds. large-sample surveys for deuterated molecules are needed to understand the enhancement of deuterated molecules from diffuse molecular gas to cold cores. A single-pointing survey toward the 559 Planck cold clumps of the Early Cold Core Catalogue (ECC) has been conducted using the Arizona Radio Observatory 12-meter telescope, focusing on the $J$=1-0 transitions of DCO$^+$ and DCN. The survey included observations of 309 cores for DCO$^+$ and DCN 1-0 simultaneously, followed by 71 of these cores where DCO$^+$ 1-0 was detected for H$^{13}$CO$^+$ and H$^{13}$CN 1-0 simultaneously, aiming to determine the deuterated fraction ($D_{\rm frac}$). Additionally, 250 cores were observed for DCO$^+$, DCN, H$^{13}$CO$^+$ and H$^{13}$CN 1-0 simultaneously. Among the 309 sources, DCO$^+$ and DCN 1-0 were detected in 79 and 11 sources, with a detection rates of 25.6 % and 3.6 % respectively. In the 250 sources observed for all four species, DCO$^+$, DCN, H$^{13}$CO$^+$ and H$^{13}$CN 1-0 were detected in 58, 9, 57 and 13 sources, with a detection rate of 23.2 %, 3.6 %, 22.8 % and 5.2 % respectively. The $D_{\rm frac}$(HCO$^+$) values in 112 sources range from 0.89 % to 7.4 % with a median value of 3.1 %, while $D_{\rm frac}$(HCN) values in 11 sources range from 1.5 % to 5.5 % with a median value of 2.3 %. The line widths of DCO$^+$ and H$^{13}$CO$^+$ 1-0 detections are mostly within 1 km s$^{-1}$. The similarity in $D_{\rm frac}$ values between HCO$^+$ and HCN indicates that the higher detection rate of DCO$^+$ 1-0 compared with DCN 1-0 is due to the lower critical density of DCO$^+$ 1-0. We suggest that the enhancement of DCO$^+$ and DCN likely begins in the early diffuse stage of the molecular cloud, rather than during the cold core formation stage.

astro-ph.GA↗

arXiv2025

Seaweed-7B: Cost-Effective Training of Video Generation Foundation Model

This technical report presents a cost-efficient strategy for training a video generation foundation model. We present a mid-sized research model with approximately 7 billion parameters (7B) called Seaweed-7B trained from scratch using 665,000 H100 GPU hours. Despite being trained with moderate computational resources, Seaweed-7B demonstrates highly competitive performance compared to contemporary video generation models of much larger size. Design choices are especially crucial in a resource-constrained setting. This technical report highlights the key design decisions that enhance the performance of the medium-sized diffusion model. Empirically, we make two observations: (1) Seaweed-7B achieves performance comparable to, or even surpasses, larger models trained on substantially greater GPU resources, and (2) our model, which exhibits strong generalization ability, can be effectively adapted across a wide range of downstream applications either by lightweight fine-tuning or continue training. See the project page at https://seaweed.video/

cs.CV↗

arXiv2025

Optimizing LLM Queries in Relational Data Analytics Workloads

Batch data analytics is a growing application for Large Language Models (LLMs). LLMs enable users to perform a wide range of natural language tasks, such as classification, entity extraction, and translation, over large datasets. However, LLM inference is highly costly and slow: for example, an NVIDIA L4 GPU running Llama3-8B can only process 6 KB of text per second, taking about a day to handle 15 GB of data; processing a similar amount of data costs around $10K on OpenAI's GPT-4o. In this paper, we propose novel techniques that can significantly reduce the cost of LLM calls for relational data analytics workloads. Our key contribution is developing efficient algorithms for reordering the rows and the fields within each row of an input table to maximize key-value (KV) cache reuse when performing LLM serving. As such, our approach can be easily applied to existing analytics systems and serving platforms. Our evaluation shows that our solution can yield up to 3.4x improvement in job completion time on a benchmark of diverse LLM-based queries using Llama 3 models. Our solution also achieves a 32% cost savings under OpenAI and Anthropic pricing models.

cs.LG↗

arXiv2025

Physical Properties and HI-to-H2 Transition Across Taurus Linear Edge

Studying the atomic-to-molecular transition is essential for understanding the evolution of interstellar medium. The linear edge of Taurus molecular cloud, clearly identified in the $^{13}$CO(1-0) intensity map, serves as an ideal site for investigating this transition. Utilizing the Arizona Radio Observatory Sub-Millimeter Telescope, we obtained mapping observations of CO(2-1), $^{13}$CO(2-1), and CO(3-2) across this linear edge. The intensity ratio between CO(2-1) and $^{13}$CO(2-1) indicates a lower limit of the $^{12}C/^{13}C$ ratio of $54\pm 17$. Based on multi-transition observations of CO and $^{13}$CO, we performed Markov Chain Monte Carlo (MCMC) fit of the physical properties across this edge using non-Local Thermodynamic Equilibrium analysis with the RADEX code, based on the Large velocity Gradient (LVG) assumption. The number density profile exhibits a pronounced jump coinciding with the H$_2$ infrared emission peak. The cold HI gas within the molecular cloud, manifested as HI-Narrow Self-Absorption (HINSA) features, is detected along the cloud edge. Our quantitative comparison with numerical simulations provides tentative evidence that shocks induced by colliding gas flows may contribute to the atomic-to-molecular phase transition observed along the linear edge.

astro-ph.GA↗

arXiv2025

Numerical analysis of a first-order computational algorithm for reaction-diffusion equations via the primal-dual hybrid gradient method

In arXiv:2305.03945 [math.NA], a first-order optimization algorithm has been introduced to solve time-implicit schemes of reaction-diffusion equations. In this research, we conduct theoretical studies on this first-order algorithm equipped with a quadratic regularization term. We provide sufficient conditions under which the proposed algorithm and its time-continuous limit converge exponentially fast to a desired time-implicit numerical solution. We show both theoretically and numerically that the convergence rate is independent of the grid size, which makes our method suitable for large-scale problems. The efficiency of our algorithm has been verified via a series of numerical examples conducted on various types of reaction-diffusion equations. The choice of optimal hyperparameters as well as comparisons with some classical root-finding algorithms are also discussed in the numerical section.

math.NA↗

arXiv2025

STEVE: A Step Verification Pipeline for Computer-use Agent Training

Developing AI agents to autonomously manipulate graphical user interfaces is a long challenging task. Recent advances in data scaling law inspire us to train computer-use agents with a scaled instruction set, yet using behavior cloning to train agents still requires immense high-quality trajectories. To meet the scalability need, we designed STEVE, a step verification pipeline for computer-use agent training. First, we establish a large instruction set for computer-use agents and collect trajectory data with some suboptimal agents. GPT-4o is used to verify the correctness of each step in the trajectories based on the screens before and after the action execution, assigning each step with a binary label. Last, we adopt the Kahneman and Tversky Optimization to optimize the agent from the binary stepwise labels. Extensive experiments manifest that our agent outperforms supervised finetuning by leveraging both positive and negative actions within a trajectory. Also, STEVE enables us to train a 7B vision-language model as a computer-use agent, achieving leading performance in the challenging live desktop environment WinAgentArena with great efficiency at a reduced cost. Code and data: https://github.com/FanbinLu/STEVE.

cs.CV↗

arXiv2025

Jenga: Effective Memory Management for Serving LLM with Heterogeneity

Large language models (LLMs) are widely used but expensive to run, especially as inference workloads grow. To lower costs, maximizing the request batch size by managing GPU memory efficiently is crucial. While PagedAttention has recently been proposed to improve the efficiency of memory management, we find that the growing heterogeneity in the embeddings dimensions, attention, and access patterns of modern LLM architectures introduces new challenges for memory allocation. In this paper, we present Jenga, a novel memory allocation framework for heterogeneous embeddings in LLMs. Jenga tackles two key challenges: (1) minimizing memory fragmentation when managing embeddings of different sizes, and (2) enabling flexible caching and eviction policies tailored to the specific token-dependency patterns of various layers. Jenga employs a two-level memory allocator, leveraging the least common multiple (LCM) of embedding sizes to optimize memory usage and providing APIs to express layer-specific caching logic to enhance memory reuse. We implemente Jenga on vLLM, a state-of-the-art LLM inference engine, and evaluate it with diverse LLMs, datasets, and GPU configurations. Evaluations show that Jenga improves GPU memory utilization by up to 79.6%, and increases serving throughput by up to 4.92x (1.80x on average).

cs.DC↗

arXiv2025

Fùxì: A Benchmark for Evaluating Language Models on Ancient Chinese Text Understanding and Generation

Ancient Chinese text processing presents unique challenges for large language models (LLMs) due to its distinct linguistic features, complex structural constraints, and rich cultural context. While existing benchmarks have primarily focused on evaluating comprehension through multiple-choice questions, there remains a critical gap in assessing models' generative capabilities in classical Chinese. We introduce Fùxì, a comprehensive benchmark that evaluates both understanding and generation capabilities across 21 diverse tasks. Our benchmark distinguishes itself through three key contributions: (1) balanced coverage of both comprehension and generation tasks, including novel tasks like poetry composition and couplet completion, (2) specialized evaluation metrics designed specifically for classical Chinese text generation, combining rule-based verification with fine-tuned LLM evaluators, and (3) a systematic assessment framework that considers both linguistic accuracy and cultural authenticity. Through extensive evaluation of state-of-the-art LLMs, we reveal significant performance gaps between understanding and generation tasks, with models achieving promising results in comprehension but struggling considerably in generation tasks, particularly those requiring deep cultural knowledge and adherence to classical formats. Our findings highlight the current limitations in ancient Chinese text processing and provide insights for future model development. The benchmark, evaluation toolkit, and baseline results are publicly available to facilitate research in this domain.

cs.CL↗

arXiv2025

Quasi-linear time decoding of RS and AG codes for burst errors up to the Singleton bound

Despite of tremendous research on decoding Reed-Solomon (RS) and algebraic geometry (AG) codes under the random and adversary substitution error models, few studies have explored these codes under the burst substitution error model. Burst errors are prevalent in many communication channels, such as wireless networks, magnetic recording systems, and flash memory. Compared to random and adversarial errors, burst errors often allow for the design of more efficient decoding algorithms. However, achieving both an optimal decoding radius and quasi-linear time complexity for burst error correction remains a significant challenge. The goal of this paper is to design (both list and probabilistic unique) decoding algorithms for RS and AG codes that achieve the Singleton bound for decoding radius while maintaining quasi-linear time complexity. Our idea is to build a one-to-one correspondence between AG codes (including RS codes) and interleaved RS codes with shorter code lengths (or even constant lengths). By decoding the interleaved RS codes with burst errors, we derive efficient decoding algorithms for RS and AG codes. For decoding interleaved RS codes with shorter code lengths, we can employ either the naive methods or existing algorithms. This one-to-one correspondence is constructed using the generalized fast Fourier transform (G-FFT) proposed by Li and Xing (SODA 2024). The G-FFT generalizes the divide-and-conquer technique from polynomials to algebraic function fields. More precisely speaking, assume that our AG code is defined over a function field $E$ which has a sequence of subfields $\mathbb{F}_q(x)=E_r\subseteq E_{r-1}\subseteq \cdots\subset E_1\subseteq E_0=E$ such that $E_{i-1}/E_i$ are Galois extensions for $1\le i\le r$. Then the AG code based on $E$ can be transformed into an interleaved RS code over the rational function field $\mathbb{F}_q(x)$.

cs.IT↗

arXiv2025

SkyStore: Cost-Optimized Object Storage Across Regions and Clouds

Modern applications span multiple clouds to reduce costs, avoid vendor lock-in, and leverage low-availability resources in another cloud. However, standard object stores operate within a single cloud, forcing users to manually manage data placement across clouds, i.e., navigate their diverse APIs and handle heterogeneous costs for network and storage. This is often a complex choice: users must either pay to store objects in a remote cloud, or pay to transfer them over the network based on application access patterns and cloud provider cost offerings. To address this, we present SkyStore, a unified object store that addresses cost-optimal data management across regions and clouds. SkyStore introduces a virtual object and bucket API to hide the complexity of interacting with multiple clouds. At its core, SkyStore has a novel TTL-based data placement policy that dynamically replicates and evicts objects according to application access patterns while optimizing for lower cost. Our evaluation shows that across various workloads, SkyStore reduces the overall cost by up to 6x over academic baselines and commercial alternatives like AWS multi-region buckets. SkyStore also has comparable latency, and its availability and fault tolerance are on par with standard cloud offerings. We release the data and code of SkyStore at https://github.com/skyplane-project/skystore.

cs.DC↗

arXiv2025

Astronomical image denoising by self-supervised deep learning and restoration processes

Image denoising based on deep learning has witnessed significant advancements in recent years. However, existing deep learning methods lack quantitative control of the deviation or error on denoised images. The neural networks Self2Self is designed for denoising single-image, training on it and denoising itself, during which training is costly. In this work we explore training Self2Self on an astronomical image and denoising other images of the same kind, which is suitable for quickly denoising massive images in astronomy. To address the deviation issue, the abnormal pixels whose deviation exceeds a predefined threshold are restored to their initial values. The noise reduction includes training, denoising, restoring and named TDR-method, by which the noise level of the solar magnetograms is improved from about 8 G to 2 G. Furthermore, the TDR-method is applied to galaxy images from the Hubble Space Telescope and makes weak galaxy structures become much clearer. This capability of enhancing weak signals makes the TDR-method applicable in various disciplines.

astro-ph.IM↗

arXiv2025

LLMs Can Easily Learn to Reason from Demonstrations Structure, not content, is what matters!

Large reasoning models (LRMs) tackle complex reasoning problems by following long chain-of-thoughts (Long CoT) that incorporate reflection, backtracking, and self-validation. However, the training techniques and data requirements to elicit Long CoT remain poorly understood. In this work, we find that a Large Language model (LLM) can effectively learn Long CoT reasoning through data-efficient supervised fine-tuning (SFT) and parameter-efficient low-rank adaptation (LoRA). With just 17k long CoT training samples, the Qwen2.5-32B-Instruct model achieves significant improvements on a wide range of math and coding benchmarks, including 56.7% (+40.0%) on AIME 2024 and 57.0% (+8.1%) on LiveCodeBench, competitive to the proprietary o1-preview model's score of 44.6% and 59.1%. More importantly, we find that the structure of Long CoT is critical to the learning process, whereas the content of individual reasoning steps has minimal impact. Perturbations affecting content, such as training on incorrect samples or removing reasoning keywords, have little impact on performance. In contrast, structural modifications that disrupt logical consistency in the Long CoT, such as shuffling or deleting reasoning steps, significantly degrade accuracy. For example, a model trained on Long CoT samples with incorrect answers still achieves only 3.2% lower accuracy compared to training with fully correct samples. These insights deepen our understanding of how to elicit reasoning capabilities in LLMs and highlight key considerations for efficiently training the next generation of reasoning models. This is the academic paper of our previous released Sky-T1-32B-Preview model. Codes are available at https://github.com/NovaSky-AI/SkyThought.

cs.AI↗

arXiv2025

The Danger of Overthinking: Examining the Reasoning-Action Dilemma in Agentic Tasks

Large Reasoning Models (LRMs) represent a breakthrough in AI problem-solving capabilities, but their effectiveness in interactive environments can be limited. This paper introduces and analyzes overthinking in LRMs. A phenomenon where models favor extended internal reasoning chains over environmental interaction. Through experiments on software engineering tasks using SWE Bench Verified, we observe three recurring patterns: Analysis Paralysis, Rogue Actions, and Premature Disengagement. We propose a framework to study these behaviors, which correlates with human expert assessments, and analyze 4018 trajectories. We observe that higher overthinking scores correlate with decreased performance, with reasoning models exhibiting stronger tendencies toward overthinking compared to non-reasoning models. Our analysis reveals that simple efforts to mitigate overthinking in agentic environments, such as selecting the solution with the lower overthinking score, can improve model performance by almost 30% while reducing computational costs by 43%. These results suggest that mitigating overthinking has strong practical implications. We suggest that by leveraging native function-calling capabilities and selective reinforcement learning overthinking tendencies could be mitigated. We also open-source our evaluation framework and dataset to facilitate research in this direction at https://github.com/AlexCuadron/Overthinking.

cs.AI↗

arXiv2025

Locality-aware Fair Scheduling in LLM Serving

Large language model (LLM) inference workload dominates a wide variety of modern AI applications, ranging from multi-turn conversation to document analysis. Balancing fairness and efficiency is critical for managing diverse client workloads with varying prefix patterns. Unfortunately, existing fair scheduling algorithms for LLM serving, such as Virtual Token Counter (VTC), fail to take prefix locality into consideration and thus suffer from poor performance. On the other hand, locality-aware scheduling algorithms in existing LLM serving frameworks tend to maximize the prefix cache hit rate without considering fair sharing among clients. This paper introduces the first locality-aware fair scheduling algorithm, Deficit Longest Prefix Match (DLPM), which can maintain a high degree of prefix locality with a fairness guarantee. We also introduce a novel algorithm, Double Deficit LPM (D$^2$LPM), extending DLPM for the distributed setup that can find a balance point among fairness, locality, and load-balancing. Our extensive evaluation demonstrates the superior performance of DLPM and D$^2$LPM in ensuring fairness while maintaining high throughput (up to 2.87$\times$ higher than VTC) and low per-client (up to 7.18$\times$ lower than state-of-the-art distributed LLM serving system) latency.

cs.DC↗

arXiv2024

Lyra: An Efficient and Speech-Centric Framework for Omni-Cognition

As Multi-modal Large Language Models (MLLMs) evolve, expanding beyond single-domain capabilities is essential to meet the demands for more versatile and efficient AI. However, previous omni-models have insufficiently explored speech, neglecting its integration with multi-modality. We introduce Lyra, an efficient MLLM that enhances multimodal abilities, including advanced long-speech comprehension, sound understanding, cross-modality efficiency, and seamless speech interaction. To achieve efficiency and speech-centric capabilities, Lyra employs three strategies: (1) leveraging existing open-source large models and a proposed multi-modality LoRA to reduce training costs and data requirements; (2) using a latent multi-modality regularizer and extractor to strengthen the relationship between speech and other modalities, thereby enhancing model performance; and (3) constructing a high-quality, extensive dataset that includes 1.5M multi-modal (language, vision, audio) data samples and 12K long speech samples, enabling Lyra to handle complex long speech inputs and achieve more robust omni-cognition. Compared to other omni-methods, Lyra achieves state-of-the-art performance on various vision-language, vision-speech, and speech-language benchmarks, while also using fewer computational resources and less training data.

cs.CV↗

arXiv2024

MoE-Lightning: High-Throughput MoE Inference on Memory-constrained GPUs

Efficient deployment of large language models, particularly Mixture of Experts (MoE), on resource-constrained platforms presents significant challenges, especially in terms of computational efficiency and memory utilization. The MoE architecture, renowned for its ability to increase model capacity without a proportional increase in inference cost, greatly reduces the token generation latency compared with dense models. However, the large model size makes MoE models inaccessible to individuals without high-end GPUs. In this paper, we propose a high-throughput MoE batch inference system, that significantly outperforms past work. MoE-Lightning introduces a novel CPU-GPU-I/O pipelining schedule, CGOPipe, with paged weights to achieve high resource utilization, and a performance model, HRM, based on a Hierarchical Roofline Model we introduce to help find policies with higher throughput than existing systems. MoE-Lightning can achieve up to 10.3x higher throughput than state-of-the-art offloading-enabled LLM inference systems for Mixtral 8x7B on a single T4 GPU (16GB). When the theoretical system throughput is bounded by the GPU memory, MoE-Lightning can reach the throughput upper bound with 2-3x less CPU memory, significantly increasing resource utilization. MoE-Lightning also supports efficient batch inference for much larger MoEs (e.g., Mixtral 8x22B and DBRX) on multiple low-cost GPUs (e.g., 2-4 T4).

cs.DC↗