Searcharxiv⌕ Search

arXiv subjects

Zhengyang Wang

Publications and source records attributed to Zhengyang Wang.

At least 19 recordsLinked to original sources
arXiv2026

FiberTune: Preserving Action-Fiber Visual Residuals in Vision-Language-Action Fine-Tuning

Action-supervised fine-tuning of vision-language-action (VLA) policies fits demonstrations effectively but constrains only the directions that change predicted actions, leaving visual structure consistent across action-equivalent states free to collapse. We formalize this as residual visual collapse along local action fibers and propose FiberTune, a training-time objective that preserves teacher-structured visual residuals without adding inference-time overhead. FiberTune uses an online action probe to estimate action-predictive feature directions, filters them from intermediate visual-token representations, and aligns the resulting probe-filtered residuals to a frozen visual teacher while regularizing their effective rank. Under identical training conditions, FiberTune improves over task-loss-only fine-tuning in every one of six controlled simulation settings spanning two benchmarks and two architectures (pi_0.5 and OpenVLA-OFT), as well as on physical SO-101 pick-place; representative gains include +10.7 percentage points SR(5) on long-horizon CALVIN ABC-to-D and physical SO-101 task success rising from 72.7% to 78.1%. Residual diagnostics show that these gains coincide with increased probe-filtered residual teacher alignment and effective rank, consistent with the action-fiber motivation.

cs.CV↗
arXiv2026

Studying Image Tokenizers as Visual Languages in Unified Multimodal Models

Image tokenizers define the ``visual language'' of unified multimodal models, yet are commonly studied through isolated metrics or generation-/understanding-only evaluations. These evaluations do not fully capture how visual tokens behave when modeled jointly with text. We build a controlled pure-autoregressive testbed and track task-specific validation losses during multimodal continual pretraining across text, image, text-to-image (T2I), and image-to-text (I2T) prediction. We examine how these losses scale and relate to downstream performance, then use them to study multimodal learnability---how well image and text tokens are jointly modeled---and tokenizer design. We find that (1) losses should be analyzed by task, since they exhibit distinct scaling behavior and rank tokenizers differently. (2) The loss--performance relationship depends on the predicted token space: for a fixed tokenizer, T2I and I2T losses correlate with generation quality, but across tokenizers, the T2I loss--performance relationship shifts with the image-token space, whereas I2T loss, computed over a shared text vocabulary, provides a more consistent signal. I2T loss also correlates with both generation and visual understanding performance after supervised finetuning. Using losses as a lens, we show that (3) better reconstruction does not necessarily yield lower task-specific losses or stronger downstream performance, and that (4) image tokenizer choice can affect text modeling under joint optimization. As case studies, we revisit three tokenizer design axes---the discriminator, semantic supervision, and vocabulary size---to examine their effects on joint modeling and downstream performance. Together, our testbed offers a complementary perspective on image tokenizers as visual languages, highlighting their interplay with text in joint multimodal training.

cs.CV↗
arXiv2026

Trait-Aware Policy Optimization for Autoregressive Multi-Trait Essay Scoring

Multi-trait essay scoring aims to provide fine-grained evaluation of writing quality across multiple dimensions. However, how to effectively post-train autoregressive scoring models remains underexplored. In this paper, we propose Trait-Aware Policy Optimization (TAPO), a post-training framework tailored to autoregressive multi-trait scoring. Our method decomposes rewards along both the sample and trait dimensions, combining global scoring consistency, trait-level accuracy, format validity, and inter-trait dependency preservation. In addition, we use enhanced prompts throughout training by incorporating original prompt texts and trait descriptions, providing richer semantic information for trait-specific score generation. Experiments across multiple backbone models show that our method consistently improves multi-trait scoring performance over supervised fine-tuning and scalar-reward optimization baselines, demonstrating the effectiveness and transferability of trait-aware post-training for essay scoring.

cs.CL↗
arXiv2026

ScienceArena: Benchmarking LLMs on Latest Scientific Olympiad Competitions

Benchmark saturation and data contamination increasingly obscure genuine scientific reasoning in frontier LLMs. We introduce \textsc{ScienceArena}, an olympiad-style benchmark from thirteen public science competitions in physics, chemistry, and biology, including IPhO and IChO 2025--2026, IBO 2023, USAPhO 2026, and USNCO 2025. Its open-ended, multi-step problems use process-credit rubrics, making faithful scoring difficult. We build ScienceArena through an expert-audited digitization pipeline that converts official exams, figures, solutions, and rubrics into structured items verified by olympiad medalists. To scale evaluation beyond costly human grading, we calibrate LLM-as-judge against medalist ground truth on archived answers from five models across IPhO and IChO; two strong judges stay within one point of expert total scores. Medalist notes show that failures often stem from visual grounding, structure fidelity, and global problem control rather than missing terminology. Evaluating fourteen recent LLMs with interleaved solving, we find that top models obtain medal-equivalent rubric scores on several public international exams, while chemistry and long-horizon consistency remain key bottlenecks. We provide an interactive \href{https://science-arena.onrender.com/}{demo}.

cs.AI↗
arXiv2026

Prefix Sliding for efficient test-time scaling

Test-time scaling uses extra test-time compute to improve performance, such as letting language models reason longer when solving a problem. As models keep the entire reasoning trace in memory via full attention, hard tasks that need long thinking can be prohibitively expensive. However, we find most intermediate reasoning tokens lose importance as the model continues reasoning. This calls into question whether retaining them is worth the cost. Based on this insight, we propose Prefix Sliding, which discards tokens during reasoning that are not part of the prefix or the window of the last few thousand tokens. The prefix has key instructions and tools available to the model, while the most recent tokens are the current reasoning the model is working on. This caps the total memory requirement regardless of how long the model reasons, allowing for efficient long-horizon test-time scaling. Without training, Prefix Sliding can make existing models 3x faster while maintaining performance. Training with Prefix Sliding using reinforcement learning can achieve better performance by enabling scaling to reasoning traces beyond a hundred thousand tokens. Ablations show Prefix Sliding outperforms summarizing intermediate tokens or vanilla sliding window. Our code is at https://github.com/Muennighoff/prefix-sliding

cs.CL↗
arXiv2026

HeaPA: Difficulty-Aware Heap Sampling and On-Policy Query Augmentation for LLM Reinforcement Learning

RLVR has become a standard recipe for training LLMs on reasoning tasks with verifiable outcomes, but when rollout generation dominates the cost, efficiency hinges on which prompts are sampled and when. In practice, prompt pools are often static or only weakly coupled to policy progress, so uniform sampling fails to track the moving capability frontier and wastes rollouts on regions that are already solved or still unreachable. Prior methods improve efficiency via filtering, curricula, adaptive rollout allocation, or teacher guidance, but they often assume a fixed pool, which does not support stable on-policy pool growth, or they introduce additional teacher cost and latency. In this work, we propose HeaPA (Heap Sampling and On-Policy Query Augmentation), which maintains a bounded, evolving pool, tracks the frontier with heap-based boundary sampling, grows the pool via on-policy augmentation under lightweight asynchronous validation, and stabilizes correlated queries via topology-aware pool statistics re-estimation and controlled reinsertion. Across two training corpora, two training recipes, and seven benchmarks, HeaPA consistently improves accuracy and reaches target performance with fewer computations at comparable wall-clock time. Analyses attribute the gains to frontier-focused sampling and on-policy pool growth, with more pronounced improvements at mid-to-large model scales. Our training code is publicly available at https://github.com/horizon-llm/HeaPA.

cs.LG↗
arXiv2026

AISE-Bench: A Full-Cycle Curated Benchmark for Information Seeking on Academic Knowledge Graphs

Large language models (LLMs) augmented with tools are emerging as autonomous agents capable of using Web engine, APIs, and code to solve complex, long-horizon tasks. Current tool-using benchmarks for information seeking on academic graphs rely on synthetic templates, simplified solution spaces, or narrow tasks such as paper-centric tasks, leaving key challenges underexplored - realistic user intent, complex multi-step API planning, rich parameter filling for APIs, grounded answers with references, and comprehensive evaluation of both the process and the outcome. We introduce AISE-Bench, a real-world, full-cycle annotated benchmark for information seeking on academic knowledge graphs. AISE-Bench release contains 1,133 QA pairs, including query taxonomies, full API execution trajectories, validated parameters, and source-grounded answers with reference links. To support high-quality annotation, we design a customized agent workflow to enable annotators to plan, execute, and revise complex API workflows efficiently. We develop a comprehensive evaluation protocol measuring answer quality, reference grounding, API-planning correctness, and execution success. Among the 14 evaluated methods, even the strongest model (PLAY2PROMPT with Gemini-3-Pro) achieves only moderate performance and often struggles with API planning and execution. AISE-Bench establishes a challenging new testbed for quantitatively evaluating and improving the stepwise correctness, grounded summarization, and traceable reasoning of multi-step API-using LLM agents. Our code and data are available at https://aise-bench.github.io/.

cs.AI↗
arXiv2026

H2HMem: A Multimodal Memory Benchmark for Agents in Human-Human Interactions

Large language model agents are increasingly deployed in human-human interaction settings, such as meeting assistants and clinical documentation systems, where they must observe conversations and retain information for downstream queries. Unlike traditional human-assistant settings, these environments are inherently multimodal, involve complex discourse phenomena such as anaphora and deixis, and contain asynchronous or conflicting information from multiple participants. However, existing memory benchmarks largely focus on single-user, text-only interactions, failing to capture these challenges. To address this gap, we introduce H2HMem, a Human-to-Human Multimodal Memory Benchmark for evaluating memory capabilities in complex human-human interactions. H2HMem includes both dyadic and multi-party conversations with multimodal information streams, and evaluates agents along three dimensions: memory recall, reasoning, and application. Experiments with advanced agents reveal substantial limitations in constructing, retaining, and utilizing memories across modalities, participants, and sessions, highlighting substantial room for improvement in next-generation LLM agents.

cs.CL↗
arXiv2026

Muon$^2$: Boosting Muon via Adaptive Second-Moment Preconditioning

Muon has emerged as a promising optimizer for large-scale foundation model pre-training by exploiting the matrix structure of neural network updates through iterative orthogonalization. However, the orthogonalization quality of Muon hinges on the number of Newton--Schulz (NS) iterations performed, which poses efficiency challenges due to its non-trivial computation and communication cost. We propose Muon$^2$, an extension of Muon, to improve both quality and efficiency by applying Adam-style adaptive second-moment preconditioning before orthogonalization. Our key insight is that the core challenge of polar approximation in Muon lies in the ill-conditioned momentum matrix, of which the spectrum is substantially improved by Muon$^2$, leading to faster convergence toward a practically sufficient orthogonalization. We further characterize the practical orthogonalization quality via directional alignment, under which Muon$^2$ demonstrates dramatic improvement over Muon at each polar step. Across GPT, LLaMA, and Mixture-of-Experts pre-training experiments up to 13B parameters, Muon$^2$ (and its memory-efficient variant Muon$^2$-F that preserves most of its benefits) consistently outperforms Muon and its variants while reducing NS iterations by 40%, and saves up to 1/4 training time over Muon when achieving the same loss.

cs.LG↗
arXiv2026

PersonaAgent: Bridging Memory and Action for Personalized LLM Agents

Large Language Model (LLM) empowered agents have recently emerged as advanced paradigms that exhibit impressive capabilities in a wide range of domains and tasks. Despite their potential, current LLM agents often adopt a one-size-fits-all approach, lacking the flexibility to respond to users' varying needs and preferences. This limitation motivates us to develop PersonaAgent, the first personalized LLM agent framework designed to address versatile personalization tasks. Specifically, PersonaAgent integrates two complementary components - a personalized memory module that includes episodic and semantic memory mechanisms; a personalized action module that enables the agent to perform tool actions tailored to the user. At the core, the persona (defined as unique system prompt for each user) functions as an intermediary: it leverages insights from personalized memory to control agent actions, while the outcomes of these actions in turn refine the memory. Based on the framework, we propose a test-time user-preference alignment strategy that simulate the latest n interactions to optimize the persona prompt, ensuring real-time user preference alignment through textual loss feedback between simulated and ground-truth responses. Experimental evaluations demonstrate that PersonaAgent significantly outperforms other baseline methods by not only personalizing the action space effectively but also scaling during test-time real-world applications. These results underscore the feasibility and potential of our approach in delivering tailored, dynamic user experiences.

cs.AI↗
arXiv2026

Harness-Bench: Measuring Harness Effects across Models in Realistic Agent Workflows

LLM agents are increasingly deployed as executable systems that use tools, modify workspaces, and produce concrete artifacts. In such workflows, performance depends not only on the base model, but also on the harness: the system layer that manages context, tools, state, constraints, permissions, tracing, and recovery. However, existing benchmarks typically abstract away execution, compare complete agent systems, or hold the harness fixed, making execution-layer variation difficult to study. We introduce Harness-Bench, a diagnostic benchmark for evaluating configuration-level harness effects in realistic agent workflows. Harness-Bench evaluates representative harness configurations across multiple model backends under shared task environments, budgets, and evaluation protocols, while preserving each harness's native execution behavior. The benchmark contains 106 sandboxed offline tasks constructed from practical agent-use patterns and manually reviewed for realism, solvability, oracle-checkability, and integrity. Each run records final artifacts, execution traces, usage statistics, and validator outputs, enabling analysis beyond final completion. Across 5,194 execution trajectories, we observe substantial variation in completion, process quality, efficiency, and failure behavior across model-harness pairings. These results suggest that agent capability should be reported at the model-harness configuration level rather than attributed to the base model alone. Our analysis further identifies recurring execution-alignment failures, where plausible reasoning becomes decoupled from tool feedback, workspace state, evidence, or verifiable output contracts. Harness-Bench provides a reproducible foundation for diagnosing and improving reliable, efficient, and auditable agent execution stacks.

cs.AI↗
arXiv2026

ReCoVer: Resilient LLM Pre-Training System via Fault-Tolerant Collective and Versatile Workload

Pre-training large language models on massive GPU clusters has made hardware faults routine rather than rare, driving the need for resilient training systems. Yet existing frameworks either focus on specific parallelism schemes or risk drifting away from a failure-free training trajectory. We propose ReCoVer, a resilient LLM pre-training system that upholds a single invariant: each iteration keeps the number of microbatches constant, ensuring per-iteration gradients remain stochastically equivalent to a failure-free run. The framework is organized as three decoupled protocol layers: (1) Fault-tolerant collectives that isolate faults from propagating across replicas; (2) in-step fine-grained recovery that preserves intra-iteration progress and prevents gradient corruption; (3) versatile-workload policy that dynamically redistributes microbatch quotas across the survivors. The design is parallelism-agnostic, integrating directly with both 3D parallelism and Hybrid Sharded Data Parallel (HSDP) as a drop-in substrate. We evaluate our implementation on end-to-end pre-training tasks for up to 512 GPUs, ReCoVer successfully preserves the training trajectory from a failure-free reference despite of 256 GPUs lost spread across the run. For comparison with checkpoint-and-restart baselines, ReCoVer demonstrates $2.23\times$ higher effective throughput after successive failures. This advantage results in ReCoVer processing 74.9% more tokens at 234 GPU-hours, with the gap widening as the training prolongs.

cs.DC↗
arXiv2026

END: Early Noise Dropping for Efficient and Effective Context Denoising

Large Language Models (LLMs) have demonstrated remarkable performance across a wide range of natural language processing tasks. However, they are often distracted by irrelevant or noisy context in input sequences that degrades output quality. This problem affects both long- and short-context scenarios, such as retrieval-augmented generation, table question-answering, and in-context learning. We reveal that LLMs can implicitly identify whether input sequences contain useful information at early layers, prior to token generation. Leveraging this insight, we introduce Early Noise Dropping (\textsc{END}), a novel approach to mitigate this issue without requiring fine-tuning the LLMs. \textsc{END} segments input sequences into chunks and employs a linear prober on the early layers of LLMs to differentiate between informative and noisy chunks. By discarding noisy chunks early in the process, \textsc{END} preserves critical information, reduces distraction, and lowers computational overhead. Extensive experiments demonstrate that \textsc{END} significantly improves both performance and efficiency across different LLMs on multiple evaluation datasets. Furthermore, by investigating LLMs' implicit understanding to the input with the prober, this work also deepens understanding of how LLMs do reasoning with contexts internally.

cs.CL↗
arXiv2026

MUON+: Towards More Effective Muon via One Additional Normalization Step for LLM Pre-training

Muon has recently emerged as a strong optimizer for large language model pre-training, orthogonalizing the momentum matrix via Newton--Schulz polar iterations. A natural intuition is that polar iterations, by flattening the singular spectrum to all ones, should also eliminate column- and row-wise norm imbalance in the update. We show that this is not true in practice: practical polar steps can substantially amplify the imbalance. We term this the post-polar imbalanced update problem, and prove that such imbalance tightens the second-order term in a blockwise descent analysis, weakening Muon's per-step descent guarantee. Motivated by this analysis, we propose Muon+, a one-line fix that inserts a single normalization step after polar orthogonalization. Muon+ adds no optimizer state. Across pre-training experiments on GPT and LLaMA models from 60M to 7B parameters, spanning both compute-optimal budgets and extended token-to-parameter ratios up to approximately 200, Muon+ consistently outperforms Muon in terms of training and validation perplexity, leading to significant overall pre-training speedup.

cs.LG↗
arXiv2026

BOOST: BOttleneck-Optimized Scalable Training Framework for Low-Rank Large Language Models

The scale of transformer model pre-training is constrained by the increasing computation and communication cost. Low-rank bottleneck architectures offer a promising solution to significantly reduce the training time and memory footprint with minimum impact on accuracy. Despite algorithmic efficiency, bottleneck architectures scale poorly under standard tensor parallelism. Simply applying 3D parallelism designed for full-rank methods leads to excessive communication and poor GPU utilization. To address this limitation, we propose BOOST, an efficient training framework tailored for large-scale low-rank bottleneck architectures. BOOST introduces a novel Bottleneck-aware Tensor Parallelism, and combines optimizations such as online-RMSNorm, linear layer grouping, and low-rank activation checkpointing to achieve end-to-end training speedup. Evaluations on different low-rank bottleneck architectures demonstrate that BOOST achieves 1.46-1.91$\times$ speedup over full-rank model baselines and 1.87-2.27$\times$ speedup over low-rank model with naively integrated 3D parallelism, with improved GPU utilization and reduced communication overhead.

cs.LG↗
arXiv2026

SkipKV: Selective Skipping of KV Generation and Storage for Efficient Inference with Large Reasoning Models

Large reasoning models (LRMs) often incur significant key-value (KV) cache overhead, due to their linear growth with the verbose chain-of-thought (CoT) reasoning. This incurs both memory overhead and throughput bottlenecks, limiting efficient deployment. To reduce KV cache size during inference, we first investigate the effectiveness of existing KV cache eviction methods for CoT reasoning. Interestingly, we find that due to unstable token-wise scoring and reduced effective KV budget caused by padding, state-of-the-art (SoTA) eviction methods fail to maintain accuracy in multi-batch settings. Additionally, these methods often generate longer sequences than the original model without eviction, as semantic-unaware token-wise eviction leads to repeated revalidation during reasoning. To address these issues, we present \textbf{SkipKV}, a \textbf{\textit{training-free}} KV compression method that performs selective \textit{eviction} and \textit{generation}, operating at a coarse-grained, sentence-level sequence removal for efficient CoT reasoning. In specific, it introduces a \textit{sentence-scoring metric} to identify and remove highly similar sentences while maintaining semantic coherence. To suppress redundant generation, SkipKV dynamically adjusts a steering vector to update the hidden activation states during inference, enforcing the LRM to generate concise responses. Extensive evaluations on multiple reasoning benchmarks demonstrate that SkipKV achieves up to $\mathbf{26.7}\%$ higher accuracy compared to baseline methods, at a similar compression budget. Additionally, compared to SoTA, SkipKV yields up to $\mathbf{1.6}\times$ shorter generation length while improving throughput by up to $\mathbf{1.7}\times$. Our code is released at: \href{https://github.com/TTTTTTris/SkipKV}{https://github.com/TTTTTTris/SkipKV}.

cs.AI↗
arXiv2026

Retrieval-Augmented Sketch-Guided 3D Building Generation

In the early design stage of Japanese detached houses, the lack of a unified design representation among clients, sales representatives, and designers leads to design drift and inefficient feedback. Usually, sketches handed off by sales representatives may lose details for quick drawing, which reduces the fidelity of subsequent 3D generation using generative AI models. The generated 3D model typically takes the form of a single unified mesh, preventing component-level editing. To solve these issues, we propose a multi-stage 3D generative design framework capable of producing architectural models from rough design sketches. The framework combines generative and retrieval-based methods to enable component-level editing and personalized customization. It adopts a multimodal representation for 3D model generation and applies component segmentation to localize architectural components such as windows and doors and uses retrieval to support targeted replacement of components. Experiments show that the work enables modular customization which is thought to be suitable for personalized architectural design. This work introduces a multi-stage sketch-to-3D framework for Japanese detached houses, provides facade and component datasets, and shows effectiveness through quantitative and expert evaluations.

cs.GR↗
arXiv2026

Sketch-Guided Stylized Landscape Cinemagraph Synthesis

Designing stylized cinemagraphs is challenging due to the difficulty in customizing complex and expressive flow elements. To achieve intuitive and detailed control of the generated cinemagraphs, sketches provide a feasible solution to convey personalized design requirements beyond text inputs. In this paper, we propose Sketch2Cinemagraph, a sketch-guided framework that enables the conditional generation of stylized cinemagraphs from freehand sketches. Sketch2Cinemagraph adopts text prompts for initial landscape generation and provides sketch controls for both spatial and motion cues. The latent diffusion model first generates target stylized landscape images along with realistic versions. Then, a pre-trained object detection model obtains masks for the flow regions. We propose a latent motion diffusion model to estimate motion field in fluid regions of the generated landscape images. The input motion sketches serve as the conditions to control the generated motion fields in the masked fluid regions with the prompt. To synthesize cinemagraph frames, the pixels within fluid regions are warped to target locations at each timestep using a U-Net based frame generator. The results verified that Sketch2Cinemagraph can generate aesthetically appealing stylized cinemagraphs with continuous temporal flow from sketch inputs. We showcase the advantages of Sketch2Cinemagraph through qualitative and quantitative comparisons against the state-of-the-art approaches.

cs.CV↗