arXiv subjects

Richeng Xuan

Publications and source records attributed to Richeng Xuan.

At least 19 recordsLinked to original sources

arXiv2026

DramaChain Bench: An End-to-End Benchmark for Short-Drama Generation

Commercial short-drama production follows a multi-stage chain: script, storyboard, keyframe imagery, shot-level video, and the finished short drama. Most existing benchmarks evaluate solely the video-generation stage using pre-authored inputs instead of real upstream pipeline outputs. This leaves two critical questions unanswerable: whether each stage adheres to the original script intent (rather than only its immediate input prompt), and whether disparate shots remain coherent after assembly into multi-episode releases. We present DramaChain Bench, the first short-drama benchmark that evaluates every stage of the complete production chain. It is built upon three in-house systems sharing one dimension system, DramaChain Dimensions: five evaluation axes instantiated at every stage, resolving into 63 leaf dimensions. DramaChain Agent is calibrated against commercial short-drama platforms in both workflow and finished short-drama quality, enabling stage-wise fair comparison across models. DramaChain Labeling System has each of the 5,785 items scored independently by three professional annotators, with all defects spatio-temporally localised and selected from a predefined defect list. This process produces 17,488 valid scores and 255,925 traceable attribution records. The human annotations confirm that upstream defects cascade across the pipeline, demonstrating that final episode quality is not governed by video generation alone. DramaChain Agentic Judge then scores every leaf dimension automatically, gathering evidence over multiple agentic rounds before judging against a per-item checklist; it reproduces the model ranking at a mean PLCC of 0.918, enough to admit new models at no annotation cost.

cs.AI↗

arXiv2026

JPO: Juris Policy Optimization for Structured Legal Reasoning in Criminal Judgment Prediction

Criminal judgment prediction requires models to infer statutory articles, charges, and sentencing outcomes from case facts. Unlike standard classification tasks, it involves a structured reasoning process in which statutes should be matched with facts, charges should be justified by statutes, and sentencing outcomes should remain consistent with charges. Existing approaches optimize final labels, and while some have attempted to evaluate reasoning quality, their evaluations are indirect, often relying on LLM-generated rubrics that reflect model-internal preferences rather than the inherent logical structure of legal adjudication. We propose Juris Policy Optimization (JPO), a post-training framework for structured legal reasoning in Chinese criminal judgment prediction. JPO first uses teacher-generated rationales to supervise a standardized four-step reasoning process, and then applies reinforcement learning with a composite reward over legal prediction quality, reasoning structure completeness, and cross-step consistency. JPO further introduces token-level advantage reweighting and adaptive clipping for legally salient reasoning segments. Experiments on multiple open-source language models and three Chinese legal benchmarks show that JPO consistently improves both judgment prediction and reasoning quality over supervised fine-tuning and reinforcement learning baselines.

cs.CL↗

arXiv2026

Modality Fault Lines: Structural Corruptions Reveal Fragile Omni-Modal Reasoning

Omni-modal large language models are increasingly evaluated on clean text--vision--audio inputs, where every channel is present, synchronized, and readily interpretable. Such scores are often taken as evidence of robust cross-modal fusion, but clean evaluation cannot tell whether success depends on stable cross-modal structure or on cues sufficient only in intact inputs. To address this gap, we define a modality fault line: a boundary at which model behavior becomes unstable when a modality remains present and human-interpretable, but its internal evidence structure is perturbed. We introduce SCEval (Structure-Corruption Evaluation) a diagnostic evaluation protocol that keeps the question, answer space, and modality channels fixed while applying controlled structural corruptions to text, vision, and audio individually and jointly. Built from $273$ human-verified tri-modal examples from Social-IQ, OmniBench, and VALOR, SCEval evaluates $15$ proprietary and open-source omni-modal systems. The results show that structural corruption lowers clean accuracy, text--vision damage forms the most stable shared fault line, and multi-modal degradation is non-additive rather than a simple function of the number of corrupted modalities. Clean omni-modal accuracy therefore does not establish that a model will remain reliable when cross-modal evidence becomes structurally unreliable.

cs.CL↗

arXiv2026

TRAJDEBUG: Tracing Error Lifecycle to Identify Critical Failures in Long-Horizon Agent Trajectories

LLM-based agentic systems have shown remarkable capabilities in complex domains, while suffering from cascading errors and difficulty in debugging. Critical error detection aims to locate the earliest error step in a failed trajectory that is responsible for the final failure. However, progress faces two main challenges. First, long trajectories make it difficult to identify individual errors, since the evidence for judging a step may be scattered across distant instructions, observations, and prior context. Second, failed trajectories often contain multiple local errors with different downstream effects, only some of which remain responsible for the final failure. In this work, we propose TrajDebug, an error-lifecycle tracing framework that addresses long-trajectory error discovery with multi-granularity history compression and evidence-based error identification, and supports critical attribution by tracing each error's resolution status and terminal impact. We further construct TrajErrBench, a benchmark of 486 manually annotated failed trajectories from Tau2Bench and SWE-Bench Pro, covering realistic tool-use and coding scenarios. Experiments across diverse agent benchmarks show that TrajDebug achieves the best overall performance over existing baselines, and application studies further demonstrate that its diagnoses provide actionable feedback for improving downstream agent success. We will release the codes and data to facilitate further research.

cs.AI↗

arXiv2026

RepoProbe: Benchmarking Architecture-Aware Repository Comprehension with Checklists

The integration of Large Language Models (LLMs) into software engineering has shifted the focus from function-level generation to repository-scale assistance. However, existing benchmarks largely rely on bug reports from GitHub Issues, which often allow models to bypass genuine understanding via pattern matching on error logs. This misalignment under-measures Edit Bias, which refers to premature generation, where models prematurely propose code modifications instead of understanding the existing repository architecture. Furthermore, current LLM-as-a-Judge scalar scoring suffers from high variance and low interpretability. This work introduces RepoProbe, a novel benchmark for evaluating repository-level code understanding through open-ended Q&A using GitHub Discussions, which focuses on open-ended architectural inquiries rather than defect reporting. To ensure rigorous evaluation, we propose a Checklist-Based Verification Protocol that decomposes answers into atomic, verifiable facts, thereby replacing subjective ratings with objective verification. Our evaluation of state-of-the-art (SOTA) LLMs reveals a persistent gap between high clarity and evidencegrounded technical correctness. It also quantitatively confirms the prevalence of edit bias, in which models prioritize code generation instead of architectural analysis. Finally, we demonstrate that our verification protocol significantly improves evaluation reliability compared to traditional evaluations with scalar scoring.

cs.SE↗

arXiv2026

Can LLM-as-a-Judge Reliably Verify Rubrics in Agentic Scenarios?

Rubric-based scoring has become a widely used paradigm in model evaluation, typically with LLM-as-a-Judge (LaaJ) for rubric scoring. However, the reliability of LaaJ for rubric scoring remains underexplored. This concern is especially pronounced in agentic scenarios, where long, complex outputs further challenge reliable scoring. To address this, we conduct a systematic meta-evaluation of LaaJ reliability for rubric verification. We introduce RuVerBench, the first benchmark for assessing LaaJ reliability in rubric verification for agentic scenarios. RuVerBench covers two prevalent agentic domains, deep research and agentic coding, with 2,458 instances, each containing a model-generated output, a rubric, and a human-annotated label indicating whether the output satisfies the rubric. Using RuVerBench, we evaluate numerous frontier LLMs and find that even the most advanced models achieve strong performance but still exhibit substantial noise. We further analyze the impact of key LaaJ strategies, including prompt design, batching, and majority voting, on rubric verification. We find that weaker models are more sensitive to prompt variations, batched verification presents a trade-off between accuracy and efficiency, and majority voting yields effective but diminishing returns. We have released our dataset and code to facilitate future research: https://github.com/THU-KEG/RuVerBench.

cs.CL↗

arXiv2026

VisualNeedle: Benchmarking Active Visual Search in Information-Dense Scenes

Frontier multimodal large language models (MLLMs) have been reported to achieve over 90% accuracy on fine-grained perception benchmarks. However, such scores do not necessarily imply faithful use of visual evidence. Prior studies have identified three shortcuts that inflate benchmark performance. First, linguistic priors and lexical cues in questions often enable models to infer plausible answers without seeing the image. Second, coarse global semantics from the visual encoder can bypass fine-grained local details. Third, in some ``think-with-images'' benchmarks, corrupting the intermediate images returned by visual tools barely affects the final answer. These findings suggest that higher input resolution or larger question pools alone do not elicit genuine active visual search. To address this, we introduce VisualNeedle, a challenging, information-dense, and fine-grained benchmark for scenes where critical evidence is spatially constrained to minute regions and not discernible at a glance. We further propose a counterfactual crop-black setting, which replaces crops returned by tools with black images of the same size, to test whether tool-enabled performance truly relies on intermediate visual evidence. We evaluate 9 promninent MLLMs across three settings: no-tool, standard tool-enabled, and crop-black. No-tool accuracy stays below 20\%, and the best tool-enabled model reaches only 56.01\%, still trailing the 63.00% human majority-vote accuracy. These results reveal persistent limitations in fine-grained visual search, while the crop-black ablation confirms that success on VisualNeedle hinges on genuine intermediate visual evidence.

cs.CV↗

arXiv2026

Trust the uncertain teacher: distilling dark knowledge via calibrated uncertainty

The core of knowledge distillation lies in transferring the teacher's rich 'dark knowledge'-subtle probabilistic patterns that reveal how classes are related and the distribution of uncertainties. While this idea is well established, teachers trained with conventional cross-entropy often fail to preserve such signals. Their distributions collapse into sharp, overconfident peaks that appear decisive but are in fact brittle, offering little beyond the hard label or subtly hindering representation-level transfer. This overconfidence is especially problematic in high-cardinality tasks, where the nuances among many plausible classes matter most for guiding a compact student. Moreover, such brittle targets reduce robustness under distribution shift, leaving students vulnerable to miscalibration in real-world conditions. To address this limitation, we revisit distillation from a distributional perspective and propose Calibrated Uncertainty Distillation (CUD), a framework designed to make dark knowledge more faithfully accessible. Instead of uncritically adopting the teacher's overconfidence, CUD encourages teachers to reveal uncertainty where it is informative and guides students to learn from targets that are calibrated rather than sharpened certainty. By directly shaping the teacher's predictive distribution before transfer, our approach balances accuracy and calibration, allowing students to benefit from both confident signals on easy cases and structured uncertainty on hard ones. Across diverse benchmarks, CUD yields students that are not only more accurate, but also more calibrated under shift and more reliable on ambiguous, long-tail inputs.

cs.LG↗

arXiv2026

How Order-Sensitive Are LLMs? OrderProbe for Deterministic Structural Reconstruction

Large language models (LLMs) excel at semantic understanding, yet their ability to reconstruct internal structure from scrambled inputs remains underexplored. Sentence-level restoration is difficult to evaluate automatically because scrambled sentences often admit multiple valid reorderings. We introduce OrderProbe, a deterministic benchmark for structural reconstruction using fixed four-character expressions in Chinese, Japanese, and Korean, which have a unique canonical order and thus support exact-match scoring. We further propose a diagnostic framework that evaluates models beyond recovery accuracy, including Semantic Accuracy, Logical Validity, Structural Consistency, Robustness, and Information Density. Experiments on twelve widely used LLMs show that structural reconstruction remains difficult even for frontier systems: zero-shot recovery frequently falls below 35%. We also observe a consistent gap between meaning-oriented generation and exact structural reconstruction, suggesting that structural robustness is not an automatic byproduct of semantic competence.

cs.CL↗

arXiv2026

QuantEval: A Benchmark for Financial Quantitative Tasks in Large Language Models

Large Language Models (LLMs) have shown strong capabilities across many domains, yet their evaluation in financial quantitative tasks remains fragmented and mostly limited to knowledge-centric question answering. We introduce QuantEval, a benchmark that evaluates LLMs across three essential dimensions of quantitative finance: knowledge-based QA, quantitative mathematical reasoning, and quantitative strategy coding. Unlike prior financial benchmarks, QuantEval integrates a CTA-style backtesting framework that executes model-generated strategies and evaluates them using financial performance metrics, enabling a more realistic assessment of quantitative coding ability. We evaluate some state-of-the-art open-source and proprietary LLMs and observe substantial gaps to human experts, particularly in reasoning and strategy coding. Finally, we conduct large-scale supervised fine-tuning and reinforcement learning experiments on domain-aligned data, demonstrating consistent improvements. We hope QuantEval will facilitate research on LLMs' quantitative finance capabilities and accelerate their practical adoption in real-world trading workflows. We additionally release the full deterministic backtesting configuration (asset universe, cost model, and metric definitions) to ensure strict reproducibility.

cs.CL↗

arXiv2025

LaoBench: A Large-Scale Multidimensional Lao Benchmark for Large Language Models

The rapid advancement of large language models (LLMs) has not been matched by their evaluation in low-resource languages, especially Southeast Asian languages like Lao. To fill this gap, we introduce \textbf{LaoBench}, the first large-scale, high-quality, and multidimensional benchmark for assessing LLM language understanding and reasoning in Lao. LaoBench contains \textbf{17,000+} expert-curated samples across three dimensions: culturally grounded knowledge application, curriculum-aligned K12 education, and bilingual translation among Lao, Chinese, and English. It includes open-source and held-out subsets, where the held-out portion enables secure black-box evaluation via a controlled service to improve fairness and data security. We construct LaoBench with a hybrid pipeline that combines expert authoring with agent-assisted verification, ensuring linguistic accuracy, cultural relevance, and educational validity. We evaluate diverse state-of-the-art open-source and closed-source LLMs, and find that even strong multilingual models lag behind human experts, particularly in culturally grounded reasoning and translation fidelity. We hope LaoBench will catalyze research on Lao and other underrepresented Southeast Asian languages for more inclusive multilingual evaluation.

cs.CL↗

arXiv2025

FlagEval Findings Report: A Preliminary Evaluation of Large Reasoning Models on Automatically Verifiable Textual and Visual Questions

We conduct a moderate-scale contamination-free (to some extent) evaluation of current large reasoning models (LRMs) with some preliminary findings. We also release ROME, our evaluation benchmark for vision language models intended to test reasoning from visual clues. We attach links to the benchmark, evaluation data, and other updates on this website: https://flageval-baai.github.io/LRM-Eval/

cs.CL↗

arXiv2025

OMHBench: Benchmarking Balanced and Grounded Omni-Modal Multi-Hop Reasoning

Multimodal Large Language Models (MLLMs) have increasingly supported omni-modal processing across text, vision, and speech. However, existing evaluation frameworks for such models suffer from critical limitations, including modality shortcuts and biased reasoning paths. To address these challenges, we propose OMHBench, a novel benchmark designed to rigorously evaluate omni-modal multi-hop reasoning. It consists of 6,144 questions with balanced reasoning paths that are jointly grounded across all three modalities. Extensive evaluation of 13 state-of-the-art models reveals that (1) a large performance gap exists between proprietary and open-source MLLMs and (2) even proprietary models exhibit high sensitivity to reasoning path variations, resulting in asymmetric omni-modal grounding. Notably, models struggle when processing the speech modality, underscoring the need for balanced, multi-hop evaluation of omni-modal intelligence.

cs.CL↗

arXiv2025

Overconfidence in LLM-as-a-Judge: Diagnosis and Confidence-Driven Solution

Large Language Models (LLMs) are widely used as automated judges, where practical value depends on both accuracy and trustworthy, risk-aware judgments. Existing approaches predominantly focus on accuracy, overlooking the necessity of well-calibrated confidence, which is vital for adaptive and reliable evaluation pipelines. In this work, we advocate a shift from accuracy-centric evaluation to confidence-driven, risk-aware LLM-as-a-Judge systems, emphasizing the necessity of well-calibrated confidence for trustworthy and adaptive evaluation. We systematically identify the Overconfidence Phenomenon in current LLM-as-a-Judges, where predicted confidence significantly overstates actual correctness, undermining reliability in practical deployment. To quantify this phenomenon, we introduce TH-Score, a novel metric measuring confidence-accuracy alignment. Furthermore, we propose LLM-as-a-Fuser, an ensemble framework that transforms LLMs into reliable, risk-aware evaluators. Extensive experiments demonstrate that our approach substantially improves calibration and enables adaptive, confidence-driven evaluation pipelines, achieving superior reliability and accuracy compared to existing baselines.

cs.AI↗

arXiv2025

FlagEvalMM: A Flexible Framework for Comprehensive Multimodal Model Evaluation

We present FlagEvalMM, an open-source evaluation framework designed to comprehensively assess multimodal models across a diverse range of vision-language understanding and generation tasks, such as visual question answering, text-to-image/video generation, and image-text retrieval. We decouple model inference from evaluation through an independent evaluation service, thus enabling flexible resource allocation and seamless integration of new tasks and models. Moreover, FlagEvalMM utilizes advanced inference acceleration tools (e.g., vLLM, SGLang) and asynchronous data loading to significantly enhance evaluation efficiency. Extensive experiments show that FlagEvalMM offers accurate and efficient insights into model strengths and limitations, making it a valuable tool for advancing multimodal research. The framework is publicly accessible at https://github.com/flageval-baai/FlagEvalMM.

cs.CV↗

arXiv2025

Memorization or Reasoning? Exploring the Idiom Understanding of LLMs

Idioms have long posed a challenge due to their unique linguistic properties, which set them apart from other common expressions. While recent studies have leveraged large language models (LLMs) to handle idioms across various tasks, e.g., idiom-containing sentence generation and idiomatic machine translation, little is known about the underlying mechanisms of idiom processing in LLMs, particularly in multilingual settings. To this end, we introduce MIDAS, a new large-scale dataset of idioms in six languages, each paired with its corresponding meaning. Leveraging this resource, we conduct a comprehensive evaluation of LLMs' idiom processing ability, identifying key factors that influence their performance. Our findings suggest that LLMs rely not only on memorization, but also adopt a hybrid approach that integrates contextual cues and reasoning, especially when processing compositional idioms. This implies that idiom understanding in LLMs emerges from an interplay between internal knowledge retrieval and reasoning-based inference.

cs.CL↗

arXiv2024

Bi-DCSpell: A Bi-directional Detector-Corrector Interactive Framework for Chinese Spelling Check

Chinese Spelling Check (CSC) aims to detect and correct potentially misspelled characters in Chinese sentences. Naturally, it involves the detection and correction subtasks, which interact with each other dynamically. Such interactions are bi-directional, i.e., the detection result would help reduce the risk of over-correction and under-correction while the knowledge learnt from correction would help prevent false detection. Current CSC approaches are of two types: correction-only or single-directional detection-to-correction interactive frameworks. Nonetheless, they overlook the bi-directional interactions between detection and correction. This paper aims to fill the gap by proposing a Bi-directional Detector-Corrector framework for CSC (Bi-DCSpell). Notably, Bi-DCSpell contains separate detection and correction encoders, followed by a novel interactive learning module facilitating bi-directional feature interactions between detection and correction to improve each other's representation learning. Extensive experimental results demonstrate a robust correction performance of Bi-DCSpell on widely used benchmarking datasets while possessing a satisfactory detection ability.

cs.CL↗

arXiv2024

HalluDial: A Large-Scale Benchmark for Automatic Dialogue-Level Hallucination Evaluation

Large Language Models (LLMs) have significantly advanced the field of Natural Language Processing (NLP), achieving remarkable performance across diverse tasks and enabling widespread real-world applications. However, LLMs are prone to hallucination, generating content that either conflicts with established knowledge or is unfaithful to the original sources. Existing hallucination benchmarks primarily focus on sentence- or passage-level hallucination detection, neglecting dialogue-level evaluation, hallucination localization, and rationale provision. They also predominantly target factuality hallucinations while underestimating faithfulness hallucinations, often relying on labor-intensive or non-specialized evaluators. To address these limitations, we propose HalluDial, the first comprehensive large-scale benchmark for automatic dialogue-level hallucination evaluation. HalluDial encompasses both spontaneous and induced hallucination scenarios, covering factuality and faithfulness hallucinations. The benchmark includes 4,094 dialogues with a total of 146,856 samples. Leveraging HalluDial, we conduct a comprehensive meta-evaluation of LLMs' hallucination evaluation capabilities in information-seeking dialogues and introduce a specialized judge language model, HalluJudge. The high data quality of HalluDial enables HalluJudge to achieve superior or competitive performance in hallucination evaluation, facilitating the automatic assessment of dialogue-level hallucinations in LLMs and providing valuable insights into this phenomenon. The dataset and the code are available at https://github.com/FlagOpen/HalluDial.

cs.CL↗