arXiv subjects

Zhengyang Wang

Publications and source records attributed to Zhengyang Wang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Beyond Parameter Arithmetic: Sparse Complementary Fusion for Distribution-Aware Model Merging

Model merging has emerged as a promising paradigm for composing the capabilities of large language models by directly operating in weight space, enabling the integration of specialized models without costly retraining. However, existing merging methods largely rely on parameter-space heuristics, which often introduce severe interference, leading to degraded generalization and unstable generation behaviors such as repetition and incoherent outputs. In this work, we propose Sparse Complementary Fusion with reverse KL (SCF-RKL), a novel model merging framework that explicitly controls functional interference through sparse, distribution-aware updates. Instead of assuming linear additivity in parameter space, SCF-RKL measures the functional divergence between models using reverse Kullback-Leibler divergence and selectively incorporates complementary parameters. This mode-seeking, sparsity-inducing design effectively preserves stable representations while integrating new capabilities. We evaluate SCF-RKL across a wide range of model scales and architectures, covering both reasoning-focused and instruction-tuned models. Extensive experiments on 24 benchmarks spanning advanced reasoning, general reasoning and knowledge, instruction following, and safety demonstrate, vision classification that SCF-RKL consistently outperforms existing model merging methods while maintaining strong generalization and generation stability.

cs.AI↗

arXiv2026

CFT-RAG: An Entity Tree Based Retrieval Augmented Generation Algorithm With Cuckoo Filter

Although retrieval-augmented generation(RAG) significantly improves generation quality by retrieving external knowledge bases and integrating generated content, it faces computational efficiency bottlenecks, particularly in knowledge retrieval tasks involving hierarchical structures for Tree-RAG. This paper proposes a Tree-RAG acceleration method based on the improved Cuckoo Filter, which optimizes entity localization during the retrieval process to achieve significant performance improvements. Tree-RAG effectively organizes entities through the introduction of a hierarchical tree structure, while the Cuckoo Filter serves as an efficient data structure that supports rapid membership queries and dynamic updates. The experiment results demonstrate that our method is much faster than naive Tree-RAG while maintaining high levels of generative quality. When the number of trees is large, our method is hundreds of times faster than naive Tree-RAG. Our work is available at https://github.com/TUPYP7180/CFT-RAG-2025.

cs.LG↗

arXiv2026

TEON: Tensorized Orthonormalization Beyond Layer-Wise Muon for Large Language Model Pre-Training

The Muon optimizer has demonstrated strong empirical performance in pre-training large language models by performing matrix-level gradient (or momentum) orthogonalization in each layer independently. In this work, we propose TEON, a principled generalization of Muon that extends orthogonalization beyond individual layers by modeling the gradients of a neural network as a structured higher-order tensor. We present TEON's improved convergence guarantee over layer-wise Muon, and further develop a practical instantiation of TEON based on the theoretical analysis with corresponding ablation. We evaluate our approach on two widely adopted architectures: GPT-style models, ranging from 130M to 774M parameters, and LLaMA-style models, ranging from 60M to 1B parameters. Experimental results show that TEON consistently improves training and validation perplexity across model scales and exhibits strong robustness under various approximate SVD schemes.

cs.LG↗

arXiv2026

Sketch-Based Facade Renovation With Generative AI: A Streamlined Framework for Bypassing As-Built Modelling in Industrial Adaptive Reuse

Facade renovation offers a more sustainable alternative to full demolition, yet producing design proposals that preserve existing structures while expressing new intent remains challenging. Current workflows typically require detailed as-built modelling before design, which is time-consuming, labour-intensive, and often involves repeated revisions. To solve this issue, we propose a three-stage framework combining generative artificial intelligence (AI) and vision-language models (VLM) that directly processes rough structural sketch and textual descriptions to produce consistent renovation proposals. First, the input sketch is used by a fine-tuned VLM model to predict bounding boxes specifying where modifications are needed and which components should be added. Next, a stable diffusion model generates detailed sketches of new elements, which are merged with the original outline through a generative inpainting pipeline. Finally, ControlNet is employed to refine the result into a photorealistic image. Experiments on datasets and real industrial buildings indicate that the proposed framework can generate renovation proposals that preserve the original structure while improving facade detail quality. This approach effectively bypasses the need for detailed as-built modelling, enabling architects to rapidly explore design alternatives, iterate on early-stage concepts, and communicate renovation intentions with greater clarity.

cs.AI↗

arXiv2025

LaX: Boosting Low-Rank Training of Foundation Models via Latent Crossing

Training foundation models such as ViTs and LLMs requires tremendous computing cost. Low-rank matrix or tensor factorization offers a parameter-efficient alternative, but often downgrades performance due to the restricted parameter space. In this work, we introduce {\textbf{Latent Crossing (LaX)}} -- a simple yet effective plug-and-play module that enhances the capacity of low-rank models by enabling information flow across low-rank subspaces. We extensively validate the benefits of LaX on pre-training tasks with ViT-Base/Large and LLaMA-like models ranging from 60M to 1B parameters. LaX boosts low-rank model performance to match or exceed the full-rank baselines while using 2-3$\times$ fewer parameters. When equipped with low-rank adapters (i.e., LoRA) for fine-tuning LLaMA-7/13B, LaX consistently improves performance on arithmetic and common sense reasoning tasks with negligible cost.

cs.LG↗

arXiv2025

CoLA: Compute-Efficient Pre-Training of LLMs via Low-Rank Activation

The full-size MLPs and the projection layers in attention introduce tremendous model sizes of large language models (LLMs), consuming extensive computational resources in pre-training. We empirically observe that the activations of pre-trained LLMs exhibit low-rank property. Motivated by such observations, we propose CoLA and its memory-efficient implementation, CoLA-M, to replace these full-size layers with compute-efficient auto-encoders that naturally enforce low-rank activations throughout training. This fundamental architectural change eliminates the activation redundancy and significantly boosts model capacity and training efficiency. Experiments on LLaMA models with 60 million to 7 billion parameters show that CoLA reduces the computing cost by $\bf 2\pmb{\times}$ and improves training throughput by $\bf 1.86\pmb{\times}$ while maintaining full-rank level performance. CoLA-M further squeezes memory cost without sacrificing throughput, offering a pre-training approach with collectively superior parameter, computing, and memory efficiency. The LLMs produced are also $\bf 2\pmb{\times}$ smaller, enabling faster inference with lower memory cost on resource-constrained platforms.

cs.LG↗

arXiv2025

Improving Large Language Models Function Calling and Interpretability via Guided-Structured Templates

Large language models (LLMs) have demonstrated strong reasoning and tool-use capabilities, yet they often fail in real-world tool-interactions due to incorrect parameterization, poor tool selection, or misinterpretation of user intent. These issues often stem from an incomplete understanding of user goals and inadequate comprehension of tool documentation. While Chain-of-Thought (CoT) prompting has proven effective for enhancing reasoning in general contexts, our analysis reveals that free-form CoT is insufficient and sometimes counterproductive for structured function-calling tasks. To address this, we introduce a curriculum-inspired framework that leverages structured reasoning templates to guide LLMs through more deliberate step-by-step instructions for generating function callings. Experimental results show that our method reduces tool-use errors, achieving 3-12% relative improvements over strong baselines across diverse model series and approaches. Moreover, our framework enhances the robustness, interpretability, and transparency of tool-using agents, advancing the development of more reliable AI assistants for real-world applications.

cs.AI↗

arXiv2025

Can Language Models Follow Multiple Turns of Entangled Instructions?

Despite significant achievements in improving the instruction-following capabilities of large language models (LLMs), the ability to process multiple potentially entangled or conflicting instructions remains a considerable challenge. Real-world scenarios often require consistency across multiple instructions over time, such as secret privacy, personal preferences, and prioritization, which demand sophisticated abilities to integrate multiple turns and carefully balance competing objectives when instructions intersect or conflict. This work presents a systematic investigation of LLMs' capabilities in handling multiple turns of instructions, covering three levels of difficulty: (1) retrieving information from instructions, (2) tracking and reasoning across turns, and (3) resolving conflicts among instructions. We construct MultiTurnInstruct~with $\sim$1.1K high-quality multi-turn conversations through the human-in-the-loop approach and result in nine capability categories, including statics and dynamics, reasoning, and multitasking. Our finding reveals an intriguing trade-off between different capabilities. While GPT models demonstrate superior memorization, they show reduced effectiveness in privacy-protection tasks requiring selective information withholding. Larger models exhibit stronger reasoning capabilities but still struggle with resolving conflicting instructions. Importantly, these performance gaps cannot be attributed solely to information loss, as models demonstrate strong BLEU scores on memorization tasks. Still, their attention mechanisms fail to integrate multiple related instructions effectively. These findings highlight critical areas for improvement in complex real-world tasks involving multi-turn instructions. Data and codes are released at https://github.com/Glaciohound/Multi-Turn-Instruct.

cs.CL↗

arXiv2025

CCSS: Hardware-Accelerated RTL Simulation with Fast Combinational Logic Computing and Sequential Logic Synchronization

As transistor counts in a single chip exceed tens of billions, the complexity of RTL-level simulation and verification has grown exponentially, often extending simulation campaigns to several months. In industry practice, RTL simulation is divided into two phases: functional debug and system validation. While system validation demands high simulation speed and is typically accelerated using FPGAs, functional debug relies on rapid compilation-rendering multi-core CPUs the primary choice. However, the limited simulation speed of CPUs has become a major bottleneck. To address this challenge, we propose CCSS, a scalable multi-core RTL simulation platform that achieves both fast compilation and high simulation throughput. CCSS accelerates combinational logic computation and sequential logic synchronization through specialized architecture and compilation strategies. It employs a balanced DAG partitioning method and efficient boolean computation cores for combinational logic, and adopts a low-latency network-on-chip (NoC) design to synchronize sequential states across cores efficiently. Experimental results show that CCSS delivers up to 12.9x speedup over state-of-the-art multi-core simulators.

cs.AR↗

arXiv2025

RecGPT Technical Report

Recommender systems are among the most impactful applications of artificial intelligence, serving as critical infrastructure connecting users, merchants, and platforms. However, most current industrial systems remain heavily reliant on historical co-occurrence patterns and log-fitting objectives, i.e., optimizing for past user interactions without explicitly modeling user intent. This log-fitting approach often leads to overfitting to narrow historical preferences, failing to capture users' evolving and latent interests. As a result, it reinforces filter bubbles and long-tail phenomena, ultimately harming user experience and threatening the sustainability of the whole recommendation ecosystem. To address these challenges, we rethink the overall design paradigm of recommender systems and propose RecGPT, a next-generation framework that places user intent at the center of the recommendation pipeline. By integrating large language models (LLMs) into key stages of user interest mining, item retrieval, and explanation generation, RecGPT transforms log-fitting recommendation into an intent-centric process. To effectively align general-purpose LLMs to the above domain-specific recommendation tasks at scale, RecGPT incorporates a multi-stage training paradigm, which integrates reasoning-enhanced pre-alignment and self-training evolution, guided by a Human-LLM cooperative judge system. Currently, RecGPT has been fully deployed on the Taobao App. Online experiments demonstrate that RecGPT achieves consistent performance gains across stakeholders: users benefit from increased content diversity and satisfaction, merchants and the platform gain greater exposure and conversions. These comprehensive improvement results across all stakeholders validates that LLM-driven, intent-centric design can foster a more sustainable and mutually beneficial recommendation ecosystem.

cs.IR↗

arXiv2025

UniConv: Unifying Retrieval and Response Generation for Large Language Models in Conversations

The rapid advancement of conversational search systems revolutionizes how information is accessed by enabling the multi-turn interaction between the user and the system. Existing conversational search systems are usually built with two different models. This separation restricts the system from leveraging the intrinsic knowledge of the models simultaneously, which cannot ensure the effectiveness of retrieval benefiting the generation. The existing studies for developing unified models cannot fully address the aspects of understanding conversational context, managing retrieval independently, and generating responses. In this paper, we explore how to unify dense retrieval and response generation for large language models in conversation. We conduct joint fine-tuning with different objectives and design two mechanisms to reduce the inconsistency risks while mitigating data discrepancy. The evaluations on five conversational search datasets demonstrate that our unified model can mutually improve both tasks and outperform the existing baselines.

cs.CL↗

arXiv2025

Normalization of Quaternionic Polynomials in Coordinate-Free Quaternionic Variables in Conjugate-Alternating Order

Quaternionic polynomials occur naturally in applications of quaternions in science and engineering, and normalization of quaternionic polynomials is a basic manipulation. Once a Groebner basis is certified for the defining ideal I of the quaternionic polynomial algebra, the normal form of a quaternionic polynomial can be computed by routine top reduction with respect to the Groebner basis. In the literature, a Groebner basis under the conjugate-alternating order of quaternionic variables was conjectured for I in 2013, but no readable and convincing proof was found. In this paper, we present the first readable certification of the conjectured Groebner basis. The certification is based on several novel techniques for reduction in free associative algebras, which enables to not only make reduction to S-polynomials more efficiently, but also reduce the number of S-polynomials needed for the certification.

cs.SC↗

arXiv2025

Scaling Laws for Predicting Downstream Performance in LLMs

Precise estimation of downstream performance in large language models (LLMs) prior to training is essential for guiding their development process. Scaling laws analysis utilizes the statistics of a series of significantly smaller sampling language models (LMs) to predict the performance of the target LLM. For downstream performance prediction, the critical challenge lies in the emergent abilities in LLMs that occur beyond task-specific computational thresholds. In this work, we focus on the pre-training loss as a more computation-efficient metric for performance estimation. Our two-stage approach FLP consists of first estimating a function that maps computational resources (e.g., FLOPs) to the pre-training Loss using a series of fully-converged sampling models, followed by mapping the pre-training loss to downstream task Performance using the intermediate models with emerged performance. In our experiments, this FLP solution accurately predicts the performance of LLMs with 7B and 13B parameters using a series of sampling LMs up to 3B, achieving error margins of 5% and 10%, respectively, and significantly outperforming the FLOPs-to-Performance approach. Further, we present FLP-M, a fundamental approach for performance prediction that addresses the practical need to integrate datasets from multiple sources during pre-training. FLP-M extends the power law analytical function to predict domain-specific pre-training loss based on FLOPs across data sources, and employs a two-layer neural network to model the non-linear relationship between multiple domain-specific loss and downstream performance. By utilizing a 3B LLM trained on a specific ratio and a series of smaller sampling LMs, FLP-M can effectively forecast the performance of 3B and 7B LLMs across various data mixtures for most benchmarks within 10% error margins.

cs.CL↗

arXiv2025

Multi-View Depth Consistent Image Generation Using Generative AI Models: Application on Architectural Design of University Buildings

In the early stages of architectural design, shoebox models are typically used as a simplified representation of building structures but require extensive operations to transform them into detailed designs. Generative artificial intelligence (AI) provides a promising solution to automate this transformation, but ensuring multi-view consistency remains a significant challenge. To solve this issue, we propose a novel three-stage consistent image generation framework using generative AI models to generate architectural designs from shoebox model representations. The proposed method enhances state-of-the-art image generation diffusion models to generate multi-view consistent architectural images. We employ ControlNet as the backbone and optimize it to accommodate multi-view inputs of architectural shoebox models captured from predefined perspectives. To ensure stylistic and structural consistency across multi-view images, we propose an image space loss module that incorporates style loss, structural loss and angle alignment loss. We then use depth estimation method to extract depth maps from the generated multi-view images. Finally, we use the paired data of the architectural images and depth maps as inputs to improve the multi-view consistency via the depth-aware 3D attention module. Experimental results demonstrate that the proposed framework can generate multi-view architectural images with consistent style and structural coherence from shoebox model inputs.

cs.GR↗

arXiv2025

From Architectural Sketch to Conceptual Representation: Using Structure-Aware Diffusion Model to Generate Renderings of School Buildings

Generative Artificial Intelligence (AI) has advanced rapidly, enabling the generation of renderings from architectural sketches. This progress has significantly improved the efficiency of communication and conceptual expression during the early stage of architectural design. However, generated images often lack the structural details from architects' sketches. While sketches typically emphasize the overall structure, crucial components such as windows and doors are often represented by simple lines or omitted entirely. For school buildings, it is essential to control architectural components, such as the shape and proportion of windows, as these factors directly influence the accuracy of the generated images in reflecting the architect's design intentions. To address this issue, we propose a structure-aware diffusion model for architectural image generation to refine expressing design intentions through retrieval augmentation. Our framework utilizes architectural components to enhance the generation process, addressing the details that may be lacking in the sketches. These components provide clear spatial and structural details, improving the model's ability to interpret and generate architectural details. The refined sketches, combined with text prompts, are fed into the proposed structure-aware diffusion model to generate detailed and realistic school building images. The experiment results demonstrate the effectiveness of our framework in generating architectural designs.

cs.GR↗

arXiv2025

CS-Bench: A Comprehensive Benchmark for Large Language Models towards Computer Science Mastery

Large language models (LLMs) have demonstrated significant potential in advancing various fields of research and society. However, the current community of LLMs overly focuses on benchmarks for analyzing specific foundational skills (e.g. mathematics and code generation), neglecting an all-round evaluation of the computer science field. To bridge this gap, we introduce CS-Bench, the first multilingual (English, Chinese, French, German) benchmark dedicated to evaluating the performance of LLMs in computer science. CS-Bench comprises approximately 10K meticulously curated test samples, covering 26 subfields across 4 key areas of computer science, encompassing various task forms and divisions of knowledge and reasoning. Utilizing CS-Bench, we conduct a comprehensive evaluation of over 30 mainstream LLMs, revealing the relationship between CS performance and model scales. We also quantitatively analyze the reasons for failures in existing LLMs and highlight directions for improvements, including knowledge supplementation and CS-specific reasoning. Further cross-capability experiments show a high correlation between LLMs' capabilities in computer science and their abilities in mathematics and coding. Moreover, expert LLMs specialized in mathematics and coding also demonstrate strong performances in several CS subfields. Looking ahead, we envision CS-Bench serving as a cornerstone for LLM applications in the CS field and paving new avenues in assessing LLMs' diverse reasoning capabilities. The CS-Bench data and evaluation code are available at https://github.com/csbench/csbench.

cs.CL↗

arXiv2025

Hephaestus: Improving Fundamental Agent Capabilities of Large Language Models through Continual Pre-Training

Due to the scarcity of agent-oriented pre-training data, LLM-based autonomous agents typically rely on complex prompting or extensive fine-tuning, which often fails to introduce new capabilities while preserving strong generalizability. We introduce Hephaestus-Forge, the first large-scale pre-training corpus designed to enhance the fundamental capabilities of LLM agents in API function calling, intrinsic reasoning and planning, and adapting to environmental feedback. Hephaestus-Forge comprises 103B agent-specific data encompassing 76,537 APIs, including both tool documentation to introduce knowledge of API functions and function calling trajectories to strengthen intrinsic reasoning. To explore effective training protocols, we investigate scaling laws to identify the optimal recipe in data mixing ratios. By continual pre-training on Hephaestus-Forge, Hephaestus outperforms small- to medium-scale open-source LLMs and rivals commercial LLMs on three agent benchmarks, demonstrating the effectiveness of our pre-training corpus in enhancing fundamental agentic capabilities and generalization of LLMs to new tasks or environments.

cs.CL↗

arXiv2024

Data Diversity Matters for Robust Instruction Tuning

Recent works have shown that by curating high quality and diverse instruction tuning datasets, we can significantly improve instruction-following capabilities. However, creating such datasets is difficult and most works rely on manual curation or proprietary language models. Automatic data curation is difficult as it is still not clear how we can define diversity for instruction tuning, how diversity and quality depend on one other, and how we can optimize dataset quality and diversity. To resolve these issue, we propose a new algorithm, Quality-Diversity Instruction Tuning (QDIT). QDIT provides a simple method to simultaneously control dataset diversity and quality, allowing us to conduct an in-depth study on the effect of diversity and quality on instruction tuning performance. From this study we draw two key insights (1) there is a natural tradeoff between data diversity and quality and (2) increasing data diversity significantly improves the worst case instruction following performance, therefore improving robustness. We validate the performance of QDIT on several large scale instruction tuning datasets, where we find it can substantially improve worst and average case performance compared to quality-driven data selection.

cs.CL↗