Searcharxiv⌕ Search

arXiv subjects

Hengyuan Zhang

Publications and source records attributed to Hengyuan Zhang.

At least 19 recordsLinked to original sources
arXiv2026

Find Your Optimal Teacher: Personalized Data Synthesis via Router-Guided Multi-Teacher Distillation

Training student models on synthetic data generated by strong teacher models is a promising way to distilling the capabilities of teachers. However, recent studies show that stronger models are not always optimal teachers, revealing a mismatch between teacher outputs and student learnability. To address this issue, we propose PerSyn (Personalized data Synthesis), a novel synthesis strategy that operates under a new ``Route then Generate'' paradigm to create data tailored to each student model, enabling it to learn more effectively. Specifically, PerSyn first assigns each prompt to its optimal teacher via a query-level router that jointly considers student learnability and teacher response quality. Each teacher then synthesizes data only for its assigned prompts, making the process more efficient than the conventional ``Generate then Select'' paradigm, where all teachers must generate parallel responses for the entire prompt set before constructing the final dataset. Extensive experiments across different model families and scales demonstrate that PerSyn consistently achieves superior or comparable performance to all baselines in instruct tuning and math reasoning settings. Further analysis verifies the effectiveness of PerSyn and offers extra insights to propel future research.

cs.LG↗
arXiv2026

Efficient Reasoning Exploration via State-Conditioned Latent Steering with Progress Guidance

Best-of-$N$ is a widely used inference strategy for complex reasoning, whose effectiveness depends on whether sampled candidates can cover diverse and high-quality reasoning paths. However, post-trained reasoning models often suffer from \emph{exploration collapse}, where independent rollouts repeatedly follow similar reasoning paths and limit the gains from increasing the rollout budget. Existing methods alleviate this issue by promoting broader exploration, but do not explicitly guide exploration toward continuations that make meaningful progress, resulting in limited exploration efficiency. To address this, we propose \emph{\underline{S}tate-conditioned \underline{P}rogress-guided \underline{S}teering} (SPS), a training-free latent steering framework. Specifically, SPS constructs a state-conditioned Direction Bank containing multiple progress-guided steering vectors for different prefix-state regions. During online inference, SPS retrieves a suitable steering vector based on the current prefix state and applies it at high-uncertainty transitions to guide the next reasoning step toward meaningful progress. Extensive experiments across multiple model scales and benchmarks demonstrate that SPS consistently outperforms strong baselines. Further analyses validate the effectiveness of its key designs and offer valuable insights for future research. The code is available at https://github.com/rattlesnakey/SPS.

cs.CL↗
arXiv2026

Massive Activations in Hybrid Linear Attention Large Language Models: Pre-Attention Spikes and Inter-Spike Plateaus

We present the first systematic study of Massive activations (MAs) in layer-interleaved HLA LLMs and uncover two architecture-aligned morphologies: MAs consistently spike immediately before full attention layers, forming pre-attention spikes (PAS), and can persist through intervening linear attention layers, giving rise to inter-spike plateaus (ISP). As full attention becomes denser, successive PAS become increasingly connected through ISP, ultimately recovering the stable MA morphology of full attention LLMs. We establish the recurrence of this organization across five linear attention architectures, six hybridization configurations, five data domains, and representative open-source hybrid models spanning 1.2B to 397B total parameters. Controlled pretraining of GDN-based hybrids at scales up to 1.3B shows that both morphologies emerge early and respond asymmetrically to output gating: full attention output gating strongly attenuates their absolute magnitudes without eliminating their layerwise organization, whereas removing GDN gates yields comparatively modest amplification. Mechanistically, our systematic-outlier analysis supports a shared lifecycle account governed by the timing of MA cancellation. PAS follows a localized write-sink-cancel process, while the extended persistence of ISP is consistent with delayed cancellation. At the full attention limit, this account recovers the stable MA morphology characteristic of full attention LLMs. Our code is available at https://github.com/StartLuxLabs/Massive-Activations-HLA.

cs.CL↗
arXiv2026

Unconventional Pressure Evolution of Spin-Density-Wave State in La$_{3}$Ni$_{2}$O$_{7}$

The discovery of pressure-induced high temperature superconductivity in the bilayer nickelate La$_{3}$Ni$_{2}$O$_{7}$ has raised the question of how its spin-density-wave (SDW) state evolves toward the superconducting regime. Here, we report a systematic electronic Raman study of La$_{3}$Ni$_{2}$O$_{7}$ single crystals under hydrostatic pressures up to 16.51 GPa. Both the SDW gap energy and the transition temperature $T_{\mathrm{SDW}}$ show an overall increase with pressure, while the dimensionless coupling ratio 2$Δ_{\text{SDW}}/(k_{\text{B}}T_{\text{SDW}})$ remains constant around $\sim7.5$, indicating a robust strong-coupling character of SDW state. At the same time, the Raman SDW peak broadens as pressure is applied, indicating a gradual weakening of long-range SDW order. These results reveal an unusual pressure evolution in which the SDW energy scale is enhanced while the SDW state becomes progressively less coherent, providing spectroscopic constraints on the magnetic correlations relevant to superconductivity in bilayer nickelates.

cond-mat.supr-con↗
arXiv2026

AptAvatar: Fast and Vivid Long-Form Audio-Driven Video Generation for Production-Ready Avatars

Production-ready audio-driven avatar generation requires efficient inference without sacrificing fidelity or motion expressiveness. However, existing acceleration methods often compromise quality through restrictive architectural choices, such as causal attention and short temporal horizons, or by reducing model capacity and resolution. Without such compromises, we propose AptAvatar, a 14B-parameter long-form audio-driven avatar generation framework that delivers fast and expressive inference. For efficiency in production-level applications, AptAvatar addresses the extreme two-step generation challenge. To bridge the gap between the multi-step teacher model and the two-step student model, we introduce Endpoint-Anchored Distribution Distillation. It augments vanilla distribution matching with a dedicated Anchor Score Estimator trained on the trajectory-endpoint distribution defined from a frozen pretrained 4-step bridge generator. This provides an attainable endpoint-level anchor for the evolving two-step student. To improve long-horizon consistency, we further introduce Self-Generated History Replay, which reuses cached outputs from earlier generator checkpoints as history conditions during chunk-wise training. This approximates inference-time conditioning on self-generated histories without costly online rollouts, mitigating quality degradation from accumulated history errors. Extensive experiments demonstrate that AptAvatar generates vivid 720p long-form avatar videos with only 2 NFEs, achieving a 60x speedup while preserving visual fidelity and long-horizon identity. Code is available at https://github.com/TaoLiveAIGC/AptAvatar

cs.CV↗
arXiv2026

Identifying the structure of La3Ni2O7 in the pressurized superconducting state

The crystal structure of La3Ni2O7 in its high-pressure superconducting state has been the subject of intense debate, with conflicting reports proposing orthorhombic (Amam or Fmmm) and tetragonal (I4/mmm) symmetries. Here, using high-pressure Raman spectroscopy down to 3 K, we resolve this controversy by tracking the structural evolution of La3Ni2O7 up to 32.7 GPa. Leveraging rigorous symmetry-based selection rules, we identify a single structural transition from the orthorhombic Amam phase to the Fmmm phase at ~14.5 GPa, signaled by a profound phonon renormalization. Crucially, the persistence of D2h symmetry across the transition rules out the tetragonal I4/mmm phase in the superconducting state in our measurements. The emergence of bulk superconductivity coincides precisely with this transition. Our results establish the orthorhombic Fmmm structure as the intrinsic host of superconductivity in La3Ni2O7 below 19.45 GPa, resolving a central structural controversy and providing a critical foundation for understanding the superconducting mechanism in bilayer nickelates.

cond-mat.supr-con↗
arXiv2026

Learning How to Use Tools, Not Just When: Pattern-Aware Tool-Integrated Reasoning

Tool-integrated reasoning (TIR) has become a key approach for improving large reasoning models (LRMs) on complex problems. Prior work has mainly studied when to invoke tools, while overlooking how tools are applied. We identify two common patterns: a calculator pattern that uses code for direct computation, and an algorithmic pattern that encodes problems as programs. Misaligned choices often cause failures even when reasoning is sound. We propose a two-stage framework that first builds code competence from both patterns and then aligns pattern selection with teacher preferences. Across challenging math datasets, our pattern-aware method substantially improves both code usage and accuracy, for instance raising Code@1 on MATH500 from 64.0% to 70.5% and on AIME24 from 26.7% to 50.0%. These gains highlight the effectiveness of a pattern-aware approach for tool-integrated reasoning.

cs.AI↗
arXiv2026

Modularized Reinforcement Learning on LLMs: From MDP Creation to Exploration and Learning

Reinforcement learning (RL) has become central to LLM post-training, yet the methods that dominate current pipelines, PPO and GRPO, represent only a narrow slice of what RL offers. Understanding why these methods prevail, and what alternatives exist, requires a principled examination of the design decisions that underlie any RL algorithm. This survey organizes that examination around three stages of algorithm construction. We begin with MDP creation: how the reward function, state space, action space, termination condition, and discount factor are, or could be, defined for LLM training. We then turn to exploration, covering temperature sampling, entropy regularization, intrinsic motivation, tree search, and curriculum learning. Finally, we address learning along four classical RL dimensions: model-free versus model-based, value-based versus policy-based versus actor-critic, on-policy versus off-policy, and credit assignment, including both Monte Carlo methods, which rely on full return estimates, and bootstrapping methods, which update estimates using other learned predictions. Mapping the LLM literature onto this taxonomy reveals a strikingly non-uniform distribution of research effort. Critic-free policy gradients and Monte Carlo credit assignment are densely populated, while value-based methods, off-policy actor-critic training, and bootstrapping-based credit assignment remain largely unexplored despite well-established counterparts in classical RL. These gaps represent concrete opportunities for transferring proven RL techniques to LLM training. By making these gaps explicit alongside the methods that have proven effective, this survey offers researchers in both RL and LLMs a shared framework for understanding current practice and identifying promising directions for future work.

cs.LG↗
arXiv2026

Pantheon360: Taming Digital Twin Generation via 3D-Aware 360° Video Diffusion

Generating complete digital twins from videos requires precise camera control, global scene coverage, and strict spatial-temporal consistency constraints that remain challenging for perspective video generators due to their limited field of view (FoV). Their narrow FoV forces long or multi-view trajectories, amplifying cross-view inconsistency and temporal drift. We argue that 360° video generation offers a natural solution: panoramic coverage simplifies trajectory design and provides a strong global context for maintaining coherence. We introduce Pantheon360: Taming Digital Twin Generation via 3D-Aware 360° Video Diffusion, a controllable 360° video generation framework that synthesizes high-fidelity videos from sparse 360° inputs. The key idea is an explicit 3D Cache, reconstructed from the input, which serves as a geometric scaffold for any user-defined camera path. This allows the diffusion model to focus on photorealistic texture refinement while the 3D Cache enforces global geometric consistency. Experiments show that Pantheon360 achieves superior visual quality and unmatched geometric coherence, enabling reliable and flexible 360° scene generation for downstream simulation and digital-twin applications.

cs.CV↗
arXiv2026

Attention Sink in Transformers: A Survey on Utilization, Interpretation, and Mitigation

As the foundational architecture of modern machine learning, Transformers have driven remarkable progress across diverse AI domains. Despite their transformative impact, a persistent challenge across various Transformers is Attention Sink (AS), in which a disproportionate amount of attention is focused on a small subset of specific yet uninformative tokens. AS complicates interpretability, significantly affecting the training and inference dynamics, and exacerbates issues such as hallucinations. In recent years, substantial research has been dedicated to understanding and harnessing AS. However, a comprehensive survey that systematically consolidates AS-related research and offers guidance for future advancements remains lacking. To address this gap, we present the first survey on AS, structured around three key dimensions that define the current research landscape: Fundamental Utilization, Mechanistic Interpretation, and Strategic Mitigation. Our work makes a pivotal contribution by highlighting the key concepts and main trends in the field, guiding researchers through the evolution of AS-related studies. We envision this survey as a valuable resource, empowering researchers to effectively manage AS within the current Transformer paradigm, while simultaneously inspiring innovative advancements for the next generation of Transformers. The paper list of this work is available at https://github.com/ZunhaiSu/Awesome-Attention-Sink.

cs.LG↗
arXiv2026

DanceHMR: Hand-Aware Whole-Body Human Mesh Recovery from Monocular Videos

Monocular video human mesh recovery is essential for digital humans, avatar animation, and embodied simulation, where both temporal stability and expressive whole-body motion are required. Existing video HMR methods produce coherent body motion but often overlook detailed hand articulation, while image-based whole-body methods recover SMPL-X meshes independently per frame, often leading to jittery and inaccurate hand motion. We present a temporally coherent whole-body HMR framework for challenging in-the-wild monocular videos. Our model unifies body context and part-specific hand observations through residual body-hand fusion, enabling stable body motion and detailed hand recovery within a single temporal architecture. We further introduce close-up-aware augmentation to improve robustness under upper-body framing. Experiments on whole-body and body-only benchmarks demonstrate improved hand reconstruction and competitive body accuracy. Our method also produces temporally stable and 2D-consistent SMPL-X motion in challenging real-world videos.

cs.CV↗
arXiv2026

Regulating oxygen content and superconductivity in La$_3$Ni$_2$O$_{7+δ}$

The synthesis of high-quality Ruddlesden-Popper (RP) nickelates remains challenging due to variations in oxygen content and the prevalence of intergrown RP phases. Precisely controlling the stoichiometry and characterizing the resulting physical properties are essential for understanding the mechanism of high-$T_c$ superconductivity in these materials. In this work, we synthesize a series of La$_3$Ni$_2$O$_{7+δ}$ samples with systematically controlled oxygen content and perform comprehensive structural and compositional analyses. Precise oxygen tuning enables us to tailor the microstructure, yielding a pure bilayer phase, a mixture of bilayer and hybrid single-layer-bilayer phases, and a predominantly bilayer phase containing trilayer intergrowths. High-pressure transport measurements reveal distinct superconducting transitions with contrasting $T_c$ values, corresponding to the bilayer phase, the hybrid phase, and trilayer inclusions. Notably, we find that oxygen content not only governs the phase purity$-$i.e., the presence of intergrowth phases$-$but also directly modulates the upper critical field ($H_{c2}$) of the bilayer superconductivity. By establishing a phase diagram of $T_c$ and $H_{c2}$ as functions of oxygen content in La$_3$Ni$_2$O$_{7+δ}$, this work advances synthetic control and provides new insights into the superconducting mechanism of RP nickelates.

cond-mat.supr-con↗
arXiv2026

Quantifying and Improving the Robustness of Retrieval-Augmented Language Models Against Spurious Features in Grounding Data

Robustness has become a critical attribute for the deployment of RAG systems in real-world applications. Existing research focuses on robustness to explicit noise (e.g., document semantics) but overlooks implicit noise (spurious features). Moreover, previous studies on spurious features in LLMs are limited to specific types (e.g., formats) and narrow scenarios (e.g., ICL). In this work, we identify and study spurious features in the RAG paradigm, a robustness issue caused by the sensitivity of LLMs to semantic-agnostic features. We then propose a novel framework, SURE, to empirically quantify the robustness of RALMs against spurious features. Beyond providing a comprehensive taxonomy and metrics for evaluation, the framework's data synthesis pipeline facilitates training-based strategies to improve robustness. Further analysis suggests that spurious features are a widespread and challenging problem in the field of RAG. Our code is available at https://github.com/maybenotime/RAG-SpuriousFeatures .

cs.CL↗
arXiv2026

Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models

Mechanistic Interpretability (MI) has emerged as a vital approach to demystify the opaque decision-making of Large Language Models (LLMs). However, existing reviews primarily treat MI as an observational science, summarizing analytical insights while lacking a systematic framework for actionable intervention. To bridge this gap, we present a practical survey structured around the pipeline: "Locate, Steer, and Improve." We formally categorize Localizing (diagnosis) and Steering (intervention) methods based on specific Interpretable Objects to establish a rigorous intervention protocol. Furthermore, we demonstrate how this framework enables tangible improvements in Alignment, Capability, and Efficiency, effectively operationalizing MI as an actionable methodology for model optimization. The curated paper list of this work is available at https://github.com/rattlesnakey/Awesome-Actionable-MI-Survey.

cs.CL↗
arXiv2026

Residual Decoding: Mitigating Hallucinations in Large Vision-Language Models via History-Aware Residual Guidance

Large Vision-Language Models (LVLMs) can reason from image-text inputs and perform well in various multimodal tasks. Despite this success, they are affected by language priors and often produce hallucinations. Hallucinations denote generated content that is grammatically and syntactically coherent, yet bears no match or direct relevance to visual input. To address this problem, we propose Residual Decoding (ResDec). It is a novel training-free method that uses historical information to aid decoding. The method relies on the internal implicit reasoning mechanism and token logits evolution mechanism of LVLMs to correct biases. Extensive experiments demonstrate that ResDec effectively suppresses hallucinations induced by language priors, significantly improves visual grounding, and reduces object hallucinations. In addition to mitigating hallucinations, ResDec also performs exceptionally well on comprehensive LVLM benchmarks, highlighting its broad applicability.

cs.CV↗
arXiv2026

Beyond Outliers: A Data-Free Layer-wise Mixed-Precision Quantization Approach Driven by Numerical and Structural Dual-Sensitivity

Layer-wise mixed-precision quantization (LMPQ) enables effective compression under extreme low-bit settings by allocating higher precision to sensitive layers. However, existing methods typically treat all intra-layer weight modules uniformly and rely on a single numerical property when estimating sensitivity, overlooking their distinct operational roles and structural characteristics. To address this, we propose NSDS, a novel calibration-free LMPQ framework driven by Numerical and Structural Dual-Sensitivity. Specifically, it first mechanistically decomposes each layer into distinct operational roles and quantifies their sensitivity from both numerical and structural perspectives. These dual-aspect scores are then aggregated into a unified layer-wise metric through a robust aggregation scheme based on MAD-Sigmoid and Soft-OR to guide bit allocation. Extensive experiments demonstrate that NSDS consistently achieves superior performance compared to various baselines across diverse models and downstream tasks, without relying on any calibration data.

cs.LG↗
arXiv2026

Spectroscopic Evidence of Competing Diagonal Spin Interactions and Spin Disproportionation in the Bilayer Nickelate La$_3$Ni$_2$O$_7$

A comprehensive spectroscopic map of the electronic, magnetic, and lattice excitations is presented for the bilayer nickelate La$_3$Ni$_2$O$_7$ using Raman scattering at ambient pressure. Upon entering the spin density wave state below 153 K, the $A_{1g}$ channel exhibits an abrupt electronic spectral gap with a clear isosbestic point. In contrast, the $B_{1g}$ and $B_{2g}$ channels are dominated by pronounced two-magnon (2M) excitations, representing an unambiguous signature of incipient Mottness. These 2M signals in both channels constitute direct evidence for two distinct in-plane spin exchange interactions along the Ni-O bonding and its diagonal directions. Calculations based on the spin wave theory further reveal that the 2M mode in the $B_{2g}$ channel arises from the competition between two bond-diagonal antiferromagnetic interactions mediated by nickel $d_{x^2-y^2}$ orbitals. Furthermore, emergent low-energy 2M excitations below 10 meV are found to originate from distinct, weaker spin moments, strongly supporting spin disproportionation. Simultaneously, an anomalous softening of $B_{1g}$ phonons from 280 down to 4.5 K is uncovered, suggesting the presence of an incipient lattice instability leading to checkerboard-type breathing modulations. Collectively, these findings identify a ground state of the bilayer nickelate characterized by competing bond-diagonal interactions, spin disproportionation, and an incipient lattice instability, establishing key ingredients for understanding the mechanism of nickelate superconductivity.

cond-mat.str-el↗
arXiv2026

Training LLMs for Divide-and-Conquer Reasoning Elevates Test-Time Scalability

Large language models (LLMs) have demonstrated strong reasoning capabilities through step-by-step chain-of-thought (CoT) reasoning. Nevertheless, at the limits of model capability, CoT often proves insufficient, and its strictly sequential nature constrains test-time scalability. A potential alternative is divide-and-conquer (DAC) reasoning, which decomposes a complex problem into subproblems to facilitate more effective exploration of the solution. Although promising, our analysis reveals a fundamental misalignment between general-purpose post-training and DAC-style inference, which limits the model's capacity to fully leverage this potential. To bridge this gap and fully unlock LLMs' reasoning capabilities on the most challenging tasks, we propose an end-to-end reinforcement learning (RL) framework to enhance their DAC-style reasoning capacity. At each step, the policy decomposes a problem into a group of subproblems, solves them sequentially, and addresses the original one conditioned on the subproblem solutions, with both decomposition and solution integrated into RL training. Under comparable training, our DAC-style framework endows the model with a higher performance ceiling and stronger test-time scalability, surpassing CoT by 8.6% in Pass@1 and 6.3% in Pass@32 on competition-level benchmarks.

cs.CL↗