arXiv subjects

Nuo Chen

Publications and source records attributed to Nuo Chen.

At least 73 records · Page 4Linked to original sources

arXiv2026

MHRC-Bench: A Multilingual Hardware Repository-Level Code Completion benchmark

Large language models (LLMs) have achieved strong performance on code completion tasks in general-purpose programming languages. However, existing repository-level code completion benchmarks focus almost exclusively on software code and largely overlook hardware description languages. In this work, we present \textbf{MHRC-Bench}, consisting of \textbf{MHRC-Bench-Train} and \textbf{MHRC-Bench-Eval}, the first benchmark designed for multilingual hardware code completion at the repository level. Our benchmark targets completion tasks and covers three major hardware design coding styles. Each completion target is annotated with code-structure-level and hardware-oriented semantic labels derived from concrete syntax tree analysis. We conduct a comprehensive evaluation of models on MHRC-Bench-Eval. Comprehensive evaluation results and analysis demonstrate the effectiveness of MHRC-Bench.

cs.PL↗

arXiv2026

Mitigating the Safety Alignment Tax with Null-Space Constrained Policy Optimization

As Large Language Models (LLMs) are increasingly deployed in real-world applications, it is important to ensure their behaviors align with human values, societal norms, and ethical principles. However, safety alignment under Reinforcement Learning (RL) often suffers from forgetting learned general abilities, which is also known as the alignment tax. To address this issue, we introduce Null-Space constrained Policy Optimization (NSPO), a novel RL framework for LLM safety alignment while preserving their core abilities. The safety policy gradients are geometrically projected into the null space of general tasks, thereby mitigating the safety alignment tax. In addition, we theoretically prove that NSPO preserves the model's original core capabilities, while still guaranteeing a descent direction for effective safety alignment. Extensive experiments demonstrate that NSPO outperforms existing methods by a large margin, achieving state-of-the-art safety performance without sacrificing accuracy on general tasks, including math, code, and instruction-following tasks. Notably, NSPO is data-efficient and only requires 40% of public human-annotated safety data from PKU-SafeRLHF to achieve promising safety performance, without a large amount of mixed general tasks data in existing alignment methods.

cs.LG↗

arXiv2026

EvoClinician: A Self-Evolving Agent for Multi-Turn Medical Diagnosis via Test-Time Evolutionary Learning

Prevailing medical AI operates on an unrealistic ''one-shot'' model, diagnosing from a complete patient file. However, real-world diagnosis is an iterative inquiry where Clinicians sequentially ask questions and order tests to strategically gather information while managing cost and time. To address this, we first propose Med-Inquire, a new benchmark designed to evaluate an agent's ability to perform multi-turn diagnosis. Built upon a dataset of real-world clinical cases, Med-Inquire simulates the diagnostic process by hiding a complete patient file behind specialized Patient and Examination agents. They force the agent to proactively ask questions and order tests to gather information piece by piece. To tackle the challenges posed by Med-Inquire, we then introduce EvoClinician, a self-evolving agent that learns efficient diagnostic strategies at test time. Its core is a ''Diagnose-Grade-Evolve'' loop: an Actor agent attempts a diagnosis; a Process Grader agent performs credit assignment by evaluating each action for both clinical yield and resource efficiency; finally, an Evolver agent uses this feedback to update the Actor's strategy by evolving its prompt and memory. Our experiments show EvoClinician outperforms continual learning baselines and other self-evolving agents like memory agents. The code is available at https://github.com/yf-he/EvoClinician

cs.AI↗

arXiv2026

Towards Automated Smart Contract Generation: Evaluation, Benchmarking, and Retrieval-Augmented Repair

Smart contracts, predominantly written in Solidity and deployed on blockchains such as Ethereum, are immutable after deployment, making functional correctness critical. However, existing evaluations of Solidity code generation rely largely on surface-level metrics (e.g., BLEU, CrystalBLEU) or manual inspection, which correlate poorly with functional correctness. In contrast to Python, Solidity lacks large-scale, execution-based benchmarks, limiting systematic evaluation of large language models for smart contract development. We introduce SolBench, a comprehensive benchmark and automated testing pipeline for Solidity that emphasizes functional correctness via differential fuzzing. SolBench consists of 28825 functions extracted from 7604 real-world smart contracts collected from Etherscan (genesis-2024), spanning ten application domains. We benchmark 14 diverse LLMs, covering open and closed models, 1.3B-671B parameters, and both general-purpose and code-specialized architectures. The dominant failure mode is missing critical intra-contract information, such as state variables and type definitions. Providing full-contract context improves accuracy but incurs prohibitive inference costs. To address this, we propose Retrieval-Augmented Repair (RAR), a cost-effective framework that integrates execution feedback into code repair. RAR uses compiler and runtime error messages to retrieve only the minimal contract snippets needed to correct a target function, avoiding full-context inference. This significantly reduces input length while improving functional correctness. We further analyze retrieval and repair strategies within RAR, demonstrating consistent gains in accuracy and efficiency. SolBench and RAR enable principled, execution-based evaluation and economical improvement of Solidity code generation. Dataset and code are publicly available at https://github.com/ZaoyuChen/SolBench.

cs.SE↗

arXiv2026

Accelerating Generative Recommendation via Simple Categorical User Sequence Compression

Although generative recommenders demonstrate improved performance with longer sequences, their real-time deployment is hindered by substantial computational costs. To address this challenge, we propose a simple yet effective method for compressing long-term user histories by leveraging inherent item categorical features, thereby preserving user interests while enhancing efficiency. Experiments on two large-scale datasets demonstrate that, compared to the influential HSTU model, our approach achieves up to a 6x reduction in computational cost and up to 39% higher accuracy at comparable cost (i.e., similar sequence length).

cs.IR↗

arXiv2026

Evaluating Conversational Recommender Systems via Large Language Models: A User-Centric Framework

Conversational recommender systems (CRSs) integrate both recommendation and dialogue tasks, making their evaluation uniquely challenging. Existing approaches primarily assess CRS performance by separately evaluating item recommendation and dialogue management using rule-based metrics. However, these methods fail to capture the real human experience, and they cannot draw direct conclusions about the system's overall performance. As conversational recommender systems become increasingly vital in e-commerce, social media, and customer support, the ability to evaluate both recommendation accuracy and dialogue management quality using a single metric, thereby authentically reflecting user experience, has become the principal challenge impeding progress in this field. In this work, we propose a user-centric evaluation framework based on large language models (LLMs) for CRSs, namely Conversational Recommendation Evaluator (CoRE). CoRE consists of two main components: (1) LLM-As-Evaluator. Firstly, we comprehensively summarize 12 key factors influencing user experience in CRSs and directly leverage LLM as an evaluator to assign a score to each factor. (2) Multi-Agent Debater. Secondly, we design a multi-agent debate framework with four distinct roles (common user, domain expert, linguist, and HCI expert) to discuss and synthesize the 12 evaluation factors into a unified overall performance score. Furthermore, we apply the proposed framework to evaluate four CRSs on two benchmark datasets. The experimental results show that CoRE aligns well with human evaluation in most of the 12 factors and the overall assessment. Especially, CoRE's overall evaluation scores demonstrate significantly better alignment with human feedback compared to existing rule-based metrics.

cs.IR↗

arXiv2026

Incentivizing In-depth Reasoning over Long Contexts with Process Advantage Shaping

Reinforcement Learning with Verifiable Rewards (RLVR) has proven effective in enhancing LLMs short-context reasoning, but its performance degrades in long-context scenarios that require both precise grounding and robust long-range reasoning. We identify the "almost-there" phenomenon in long-context reasoning, where trajectories are largely correct but fail at the final step, and attribute this failure to two factors: (1) the lack of high reasoning density in long-context QA data that push LLMs beyond mere grounding toward sophisticated multi-hop reasoning; and (2) the loss of valuable learning signals during long-context RL training due to the indiscriminate penalization of partially correct trajectories with incorrect outcomes. To overcome this bottleneck, we propose DeepReasonQA, a KG-driven synthesis framework that controllably constructs high-difficulty, multi-hop long-context QA pairs with inherent reasoning chains. Building on this, we introduce Long-context Process Advantage Shaping (LongPAS), a simple yet effective method that performs fine-grained credit assignment by evaluating reasoning steps along Validity and Relevance dimensions, which captures critical learning signals from "almost-there" trajectories. Experiments on three long-context reasoning benchmarks show that our approach substantially outperforms RLVR baselines and matches frontier LLMs while using far fewer parameters. Further analysis confirms the effectiveness of our methods in strengthening long-context reasoning while maintaining stable RL training.

cs.CL↗

arXiv2026

AlgBench: To What Extent Do Large Reasoning Models Understand Algorithms?

Reasoning ability has become a central focus in the advancement of Large Reasoning Models (LRMs). Although notable progress has been achieved on several reasoning benchmarks such as MATH500 and LiveCodeBench, existing benchmarks for algorithmic reasoning remain limited, failing to answer a critical question: Do LRMs truly master algorithmic reasoning? To answer this question, we propose AlgBench, an expert-curated benchmark that evaluates LRMs under an algorithm-centric paradigm. AlgBench consists of over 3,000 original problems spanning 27 algorithms, constructed by ACM algorithmic experts and organized under a comprehensive taxonomy, including Euclidean-structured, non-Euclidean-structured, non-optimized, local-optimized, global-optimized, and heuristic-optimized categories. Empirical evaluations on leading LRMs (e.g., Gemini-3-Pro, DeepSeek-v3.2-Speciale and GPT-o3) reveal substantial performance heterogeneity: while models perform well on non-optimized tasks (up to 92%), accuracy drops sharply to around 49% on globally optimized algorithms such as dynamic programming. Further analysis uncovers \textbf{strategic over-shifts}, wherein models prematurely abandon correct algorithmic designs due to necessary low-entropy tokens. These findings expose fundamental limitations of problem-centric reinforcement learning and highlight the necessity of an algorithm-centric training paradigm for robust algorithmic reasoning.

cs.AI↗

arXiv2026

MiMo-V2-Flash Technical Report

We present MiMo-V2-Flash, a Mixture-of-Experts (MoE) model with 309B total parameters and 15B active parameters, designed for fast, strong reasoning and agentic capabilities. MiMo-V2-Flash adopts a hybrid attention architecture that interleaves Sliding Window Attention (SWA) with global attention, with a 128-token sliding window under a 5:1 hybrid ratio. The model is pre-trained on 27 trillion tokens with Multi-Token Prediction (MTP), employing a native 32k context length and subsequently extended to 256k. To efficiently scale post-training compute, MiMo-V2-Flash introduces a novel Multi-Teacher On-Policy Distillation (MOPD) paradigm. In this framework, domain-specialized teachers (e.g., trained via large-scale reinforcement learning) provide dense and token-level reward, enabling the student model to perfectly master teacher expertise. MiMo-V2-Flash rivals top-tier open-weight models such as DeepSeek-V3.2 and Kimi-K2, despite using only 1/2 and 1/3 of their total parameters, respectively. During inference, by repurposing MTP as a draft model for speculative decoding, MiMo-V2-Flash achieves up to 3.6 acceptance length and 2.6x decoding speedup with three MTP layers. We open-source both the model weights and the three-layer MTP weights to foster open research and community collaboration.

cs.CL↗

arXiv2026

Judging with Personality and Confidence: A Study on Personality-Conditioned LLM Relevance Assessment

Recent studies have shown that prompting can enable large language models (LLMs) to simulate specific personality traits and produce behaviors that align with those traits. However, there is limited understanding of how these simulated personalities influence critical web search decisions, specifically relevance assessment. Moreover, few studies have examined how simulated personalities impact confidence calibration, specifically the tendencies toward overconfidence or underconfidence. This gap exists even though psychological literature suggests these biases are trait-specific, often linking high extraversion to overconfidence and high neuroticism to underconfidence. To address this gap, we conducted a comprehensive study evaluating multiple LLMs, including commercial models and open-source models, prompted to simulate Big Five personality traits. We tested these models across three test collections (TREC DL 2019, TREC DL 2020, and LLMJudge), collecting two key outputs for each query-document pair: a relevance judgment and a self-reported confidence score. The findings show that personalities such as low agreeableness consistently align more closely with human labels than the unprompted condition. Additionally, low conscientiousness performs well in balancing the suppression of both overconfidence and underconfidence. We also observe that relevance scores and confidence distributions vary systematically across different personalities. Based on the above findings, we incorporate personality-conditioned scores and confidence as features in a random forest classifier. This approach achieves performance that surpasses the best single-personality condition on a new dataset (TREC DL 2021), even with limited training data. These findings highlight that personality-derived confidence offers a complementary predictive signal, paving the way for more reliable and human-aligned LLM evaluators.

cs.CL↗

arXiv2025

MiMo-Audio: Audio Language Models are Few-Shot Learners

Existing audio language models typically rely on task-specific fine-tuning to accomplish particular audio tasks. In contrast, humans are able to generalize to new audio tasks with only a few examples or simple instructions. GPT-3 has shown that scaling next-token prediction pretraining enables strong generalization capabilities in text, and we believe this paradigm is equally applicable to the audio domain. By scaling MiMo-Audio's pretraining data to over one hundred million of hours, we observe the emergence of few-shot learning capabilities across a diverse set of audio tasks. We develop a systematic evaluation of these capabilities and find that MiMo-Audio-7B-Base achieves SOTA performance on both speech intelligence and audio understanding benchmarks among open-source models. Beyond standard metrics, MiMo-Audio-7B-Base generalizes to tasks absent from its training data, such as voice conversion, style transfer, and speech editing. MiMo-Audio-7B-Base also demonstrates powerful speech continuation capabilities, capable of generating highly realistic talk shows, recitations, livestreaming and debates. At the post-training stage, we curate a diverse instruction-tuning corpus and introduce thinking mechanisms into both audio understanding and generation. MiMo-Audio-7B-Instruct achieves open-source SOTA on audio understanding benchmarks (MMSU, MMAU, MMAR, MMAU-Pro), spoken dialogue benchmarks (Big Bench Audio, MultiChallenge Audio) and instruct-TTS evaluations, approaching or surpassing closed-source models. Model checkpoints and full evaluation suite are available at https://github.com/XiaomiMiMo/MiMo-Audio.

cs.CL↗

arXiv2025

Reductive Contact and Dipolar Interface Engineering Enable Stable Flexible CsSnI3 Nanowire Photodetectors

Lead-free tin-based halide perovskites are attractive for flexible and environmentally benign optoelectronics, but their application is limited by the rapid oxidation of Sn2+ to Sn4+ and poor operational stability. Here, we report a flexible CsSnI3 nanowire photodetector that achieves both high near-infrared photoresponse and long-term stability through synergistic aluminium-substrate contact engineering and dipolar interface modification. A 0.2 mm anodized aluminium foil serves as the flexible substrate, where localized laser ablation exposes metallic aluminium regions that act as reductive sites, effectively suppressing Sn2+ oxidation during nanowire growth. Simultaneously, a polar interlayer of 3-fluoro-2-nitroanisole is introduced to improve energy-level alignment, suppress interfacial deprotonation, and enhance charge extraction. The resulting device exhibits a responsivity of 0.39 A W-1, a specific detectivity of 1.38 * 10^13 Jones, and a wide linear dynamic range of 156 dB under 850 nm illumination. Moreover, the device retains over 85% of its initial photocurrent after 60 days in ambient air and maintains 94% of its initial photocurrent after 1000 bending cycles. This work establishes an effective strategy for stabilizing Sn-based perovskites toward high-performance flexible optoelectronic devices.

cond-mat.mtrl-sci↗

arXiv2025

Soliton-Assisted Massive Signal Broadcasting via Exceptional Points

Chip-scale all-optical signal broadcasting enables data replication from an optical signal to a large number of wavelength channels, playing a critical role in enabling massive-throughput optical communication and computing systems. The underlying process is four-wave mixing between an optical signal and a multi-wavelength pump source via optical Kerr nonlinearity. To enhance the generally weak nonlinearity, high-quality (Q) microcavities are commonly used to achieve practical efficiency. However, the ultra-narrow linewidths of high Q cavities prohibit achieving massive throughput broadcasting due to Fourier reciprocity. Here, we overcome this challenge by harnessing a parity-time symmetric coupled-cavity system that supports equally spaced exceptional points in the frequency domain. This design seamlessly integrates generation of dissipative Kerr soliton comb source and all-optical signal broadcasting into a unified nonlinear process. As a result, we realize soliton-assisted intracavity massive signal broadcasting with a channel count exceeding 100 over 200 nm wavelength range, resulting in Terabit-per-second aggregated rates. This throughput surpasses the intrinsic microcavity linewidth constraint (~200 MHz) by over three orders of magnitude. We further demonstrate the utility of this approach through an optical convolutional accelerator, highlighting its potential to enable transformative capabilities in photonic computing. Our work establishes a new paradigm for chip-scale photonic processing devices based on non-Hermitian optical design.

physics.optics↗

arXiv2025

HLStrans: Dataset for C-to-HLS Hardware Code Synthesis

High-Level Synthesis (HLS) enables hardware design from C/C++ kernels but requires extensive transformations, such as restructuring code, inserting pragmas, adapting data types, and repairing non-synthesizable constructs, to achieve efficient FPGA implementations. While large language models (LLMs) show promise in automating these transformations, progress has been limited by the absence of large-scale, well-structured datasets. Existing HLS datasets focus primarily on resource estimation, lack paired C and HLS examples with testbenches, and cover only a narrow set of optimizations. We introduce HLStrans, the first benchmark-scale dataset for LLM-driven C-to-HLS synthesis. HLStrans contains over 124K paired C and HLS programs for real-world applications, with full testbenches and synthesis-based annotations of latency and resource usage. The dataset systematically captures five categories of transformations and is enriched by an automated augmentation pipeline combining LLMs, Monte Carlo Tree Search (MCTS), and Design Space Exploration (DSE). We benchmark state-of-the-art LLMs on HLStrans, demonstrating that retrieval and fine-tuning significantly improve success rates and performance.

cs.AR↗

arXiv2025

Mitigating the Threshold Priming Effect in Large Language Model-Based Relevance Judgments via Personality Infusing

Recent research has explored LLMs as scalable tools for relevance labeling, but studies indicate they are susceptible to priming effects, where prior relevance judgments influence later ones. Although psychological theories link personality traits to such biases, it is unclear whether simulated personalities in LLMs exhibit similar effects. We investigate how Big Five personality profiles in LLMs influence priming in relevance labeling, using multiple LLMs on TREC 2021 and 2022 Deep Learning Track datasets. Our results show that certain profiles, such as High Openness and Low Neuroticism, consistently reduce priming susceptibility. Additionally, the most effective personality in mitigating priming may vary across models and task types. Based on these findings, we propose personality prompting as a method to mitigate threshold priming, connecting psychological evidence with LLM-based evaluation practices.

cs.CL↗

arXiv2025

ExtendAttack: Attacking Servers of LRMs via Extending Reasoning

Large Reasoning Models (LRMs) have demonstrated promising performance in complex tasks. However, the resource-consuming reasoning processes may be exploited by attackers to maliciously occupy the resources of the servers, leading to a crash, like the DDoS attack in cyber. To this end, we propose a novel attack method on LRMs termed ExtendAttack to maliciously occupy the resources of servers by stealthily extending the reasoning processes of LRMs. Concretely, we systematically obfuscate characters within a benign prompt, transforming them into a complex, poly-base ASCII representation. This compels the model to perform a series of computationally intensive decoding sub-tasks that are deeply embedded within the semantic structure of the query itself. Extensive experiments demonstrate the effectiveness of our proposed ExtendAttack. Remarkably, it significantly increases response length and latency, with the former increasing by over 2.7 times for the o3 model on the HumanEval benchmark. Besides, it preserves the original meaning of the query and achieves comparable answer accuracy, showing the stealthiness.

cs.CR↗

arXiv2025

JudgeLRM: Large Reasoning Models as a Judge

Large Language Models (LLMs) are increasingly adopted as evaluators, offering a scalable alternative to human annotation. However, existing supervised fine-tuning (SFT) approaches often fall short in domains that demand complex reasoning. Judgment is inherently reasoning-intensive: beyond surface-level scoring, it requires verifying evidence, identifying errors, and justifying decisions. Through the analysis of evaluation tasks, we find a negative correlation between SFT performance gains and the proportion of reasoning-demanding samples, revealing the limits of SFT in such scenarios. To address this, we introduce JudgeLRM, a family of judgment-oriented LLMs, trained using reinforcement learning (RL) with judge-wise, outcome-driven rewards to activate reasoning capabilities. JudgeLRM consistently outperform SFT-tuned baselines in the same size, as well as other RL and SFT variants, and even surpass state-of-the-art reasoning models: notably, JudgeLRM-3B/4B exceeds GPT-4, while JudgeLRM-7B/8B/14B outperforms DeepSeek-R1 by over 2% in F1 score, with particularly strong gains on reasoning-heavy tasks. Our findings underscore the value of RL in unlocking reasoning-aligned LLM judges.

cs.CL↗

arXiv2025

Chain of Execution Supervision Promotes General Reasoning in Large Language Models

Building robust and general reasoning ability is a central goal in the development of large language models (LLMs). Recent efforts increasingly turn to code as a rich training source, given its inherent logical structure and diverse reasoning paradigms such as divide-and-conquer, topological ordering, and enumeration. However, reasoning in code is often expressed implicitly and entangled with syntactic or implementation noise, making direct training on raw code suboptimal.To address this, we introduce TracePile, a large-scale corpus of 2.6 million samples that transforms code execution into explicit, step-by-step chain-of-thought-style rationales, which we call Chain of Execution (CoE). The corpus spans domains including mathematics, classical algorithms and algorithmic competition, and is enriched with variable-tracing questions and code rewritings to enhance logical granularity and code diversity. We evaluate TracePile using three training setups: continue-pretraining, instruction tuning after pretraining, and two-stage finetuning. Experiments across four base models (LLaMA 3, LLaMA 3.1, Qwen-2.5, and Qwen-2.5 Coder) and 20 benchmarks covering math, code, logic, and algorithms demonstrate consistent improvements. Notably, TracePile boosts LLaMA3.1-8B by 7.1\% on average across nine math datasets and delivers clear gains on LiveCodeBench, CRUX, and MMLU under two-stage fine-tuning.

cs.LG↗