arXiv subjects

Xiaohui Hu

Publications and source records attributed to Xiaohui Hu.

At least 19 recordsLinked to original sources

arXiv2026

Prompt-Level Reward Specifications for Open-Ended Post-Training

Open-ended post-training benefits from rewards that make prompt-specific success conditions explicit, rather than relying only on post-hoc scalar scores. In instruction following, writing, and decision-support tasks, response quality depends on local requirements, holistic preferences, and explicit constraints, but existing reward methods often leave these criteria implicit or cover only narrowly verifiable cases. We propose a prompt-level reward specification framework that separates reward specification from reward computation. Given only prompts, our framework constructs reusable task-adaptive rubrics and executable hard-constraint checkers offline, making reward criteria explicit before training and reusable across rollouts. At scoring time, artifact-anchored rubric and code scores are combined with an independent global score for residual holistic quality, yielding a normalized hybrid reward over requirement satisfaction, holistic quality, and deterministic constraints. The framework requires no human preference annotations, reference answers, or a separately trained reward model. Experiments show that the resulting reward improves offline RM-style response ranking and supports online reinforcement learning across multiple open-ended benchmarks. Ablations further show that rubrics, global scoring, and executable verification provide complementary supervision.

cs.CL↗

arXiv2026

Awakening Dormant Experts:Counterfactual Routing to Mitigate MoE Hallucinations

Sparse Mixture-of-Experts (MoE) models have achieved remarkable scalability, yet they remain vulnerable to hallucinations, particularly when processing long-tail knowledge. We identify that this fragility stems from static Top-$k$ routing: routers tend to favor high-frequency patterns over rare factual associations. Consequently, ``specialist experts'' possessing critical long-tail knowledge are often assigned low gating scores and remain ``dormant'' -- under-prioritized for specific tokens despite their proven causal importance on other inputs. To address this, we propose Counterfactual Routing (CoR), a training-free inference framework designed to awaken these dormant experts. CoR integrates layer-wise perturbation analysis with the Counterfactual Expert Impact (CEI) metric to dynamically shift computational resources from syntax-dominant to knowledge-intensive layers while maintaining a constant total activation count, effectively retrieving causally decisive experts via virtual ablation. Extensive experiments on TruthfulQA, FACTOR, and TriviaQA demonstrate that CoR improves factual accuracy by 3.1\% on average without increasing the inference budget, establishing a superior Pareto frontier compared to static scaling strategies.

cs.LG↗

arXiv2026

An Effective and Cost-Efficient Agentic Framework for Ethereum Smart Contract Auditing

Smart contract security is paramount, but identifying intricate business logic vulnerabilities remains a persistent challenge because existing solutions consistently fall short: manual auditing is unscalable, static analysis tools are plagued by false positives, and fuzzers struggle to navigate deep logic states within complex systems. Even emerging AI-based methods suffer from hallucinations, context constraints, and a heavy reliance on expensive, proprietary Large Language Models. In this paper, we introduce Heimdallr, an automated auditing agent designed to overcome these hurdles through four core innovations. By reorganizing code at the function level, Heimdallr minimizes context overhead while preserving essential business logic. It then employs heuristic reasoning to detect complex vulnerabilities and automatically chain functional exploits. Finally, a cascaded verification layer validates these findings to eliminate false positives. Notably, this approach achieves high performance on lightweight, open-source models like GPToss-120B without relying on proprietary systems. Our evaluations demonstrate exceptional performance, as Heimdallr successfully reconstructed 17 out of 20 real-world attacks post June 2025, resulting in total losses of $384M, and uncovered 4 confirmed zero-day vulnerabilities that safeguarded $400M in TVL. Compared to SOTA baselines including both official industrial tools and academic tools, Heimdallr at most reduces analysis time by 97.59% and financial costs by 98.77% while boosting detection precision by over 93.66%. Notably, when applied to auditing contests, Heimdallr can achieve a 92.45% detection rate at a negligible cost of $2.31 per 10K LOC. We provide production-ready auditing services and release valuable benchmarks for future work.

cs.CR↗

arXiv2026

Beyond Binary: Turning Partial Success into Dense Verifiable Rewards for Reinforcement Learning in Code Generation

Effective reward design is a central challenge in Reinforcement Learning (RL) for code generation. Mainstream test-suite-level outcome rewards enforce functional correctness but induce sparsity, while external Reward Models (RMs) provide dense supervision at the cost of misalignment and additional overhead. Since code evaluation naturally yields multiple test-case-level outcomes, partial success, i.e., passing a subset of test cases, offers an intrinsic, verifiable source of dense supervision. In this paper, we propose VeRPO (Verifiable Dense Reward Policy Optimization), an RL framework that systematically turns verifiable partial success into reliable dense rewards. We analyze partial-success rewards using a weighted sum formulation, theoretically identifying a critical cardinality bias that causes policy updates to disproportionately favor gains from easy-test successes over progress on frontier tests. Based on this, VeRPO introduces a dynamic, density-calibrated local reward that explicitly corrects this bias and provides robust dense supervision from partial success. To enhance alignment with end-to-end functional correctness, VeRPO further integrates the local dense reward with global execution outcomes. Extensive experiments across diverse benchmarks and settings demonstrate that VeRPO outperforms outcome-driven and RM-based baselines, achieving up to +8.83 pass@1 gain with negligible time cost (< 0.02%) and zero GPU memory overhead.

cs.LG↗

arXiv2025

Training Report of TeleChat3-MoE

TeleChat3-MoE is the latest series of TeleChat large language models, featuring a Mixture-of-Experts (MoE) architecture with parameter counts ranging from 105 billion to over one trillion,trained end-to-end on Ascend NPU cluster. This technical report mainly presents the underlying training infrastructure that enables reliable and efficient scaling to frontier model sizes. We detail systematic methodologies for operator-level and end-to-end numerical accuracy verification, ensuring consistency across hardware platforms and distributed parallelism strategies. Furthermore, we introduce a suite of performance optimizations, including interleaved pipeline scheduling, attention-aware data scheduling for long-sequence training,hierarchical and overlapped communication for expert parallelism, and DVM-based operator fusion. A systematic parallelization framework, leveraging analytical estimation and integer linear programming, is also proposed to optimize multi-dimensional parallelism configurations. Additionally, we present methodological approaches to cluster-level optimizations, addressing host- and device-bound bottlenecks during large-scale training tasks. These infrastructure advancements yield significant throughput improvements and near-linear scaling on clusters comprising thousands of devices, providing a robust foundation for large-scale language model development on hardware ecosystems.

cs.CL↗

arXiv2025

Technical Report of TeleChat2, TeleChat2.5 and T1

We introduce the latest series of TeleChat models: \textbf{TeleChat2}, \textbf{TeleChat2.5}, and \textbf{T1}, offering a significant upgrade over their predecessor, TeleChat. Despite minimal changes to the model architecture, the new series achieves substantial performance gains through enhanced training strategies in both pre-training and post-training stages. The series begins with \textbf{TeleChat2}, which undergoes pretraining on 10 trillion high-quality and diverse tokens. This is followed by Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) to further enhance its capabilities. \textbf{TeleChat2.5} and \textbf{T1} expand the pipeline by incorporating a continual pretraining phase with domain-specific datasets, combined with reinforcement learning (RL) to improve performance in code generation and mathematical reasoning tasks. The \textbf{T1} variant is designed for complex reasoning, supporting long Chain-of-Thought (CoT) reasoning and demonstrating substantial improvements in mathematics and coding. In contrast, \textbf{TeleChat2.5} prioritizes speed, delivering rapid inference. Both flagship models of \textbf{T1} and \textbf{TeleChat2.5} are dense Transformer-based architectures with 115B parameters, showcasing significant advancements in reasoning and general task performance compared to the original TeleChat. Notably, \textbf{T1-115B} outperform proprietary models such as OpenAI's o1-mini and GPT-4o. We publicly release \textbf{TeleChat2}, \textbf{TeleChat2.5} and \textbf{T1}, including post-trained versions with 35B and 115B parameters, to empower developers and researchers with state-of-the-art language models tailored for diverse applications.

cs.CL↗

arXiv2025

User Long-Term Multi-Interest Retrieval Model for Recommendation

User behavior sequence modeling, which captures user interest from rich historical interactions, is pivotal for industrial recommendation systems. Despite breakthroughs in ranking-stage models capable of leveraging ultra-long behavior sequences with length scaling up to thousands, existing retrieval models remain constrained to sequences of hundreds of behaviors due to two main challenges. One is strict latency budget imposed by real-time service over large-scale candidate pool. The other is the absence of target-aware mechanisms and cross-interaction architectures, which prevent utilizing ranking-like techniques to simplify long sequence modeling. To address these limitations, we propose a new framework named User Long-term Multi-Interest Retrieval Model(ULIM), which enables thousand-scale behavior modeling in retrieval stages. ULIM includes two novel components: 1)Category-Aware Hierarchical Dual-Interest Learning partitions long behavior sequences into multiple category-aware subsequences representing multi-interest and jointly optimizes long-term and short-term interests within specific interest cluster. 2)Pointer-Enhanced Cascaded Category-to-Item Retrieval introduces Pointer-Generator Interest Network(PGIN) for next-category prediction, followed by next-item retrieval upon the top-K predicted categories. Comprehensive experiments on Taobao dataset show that ULIM achieves substantial improvement over state-of-the-art methods, and brings 5.54% clicks, 11.01% orders and 4.03% GMV lift for Taobaomiaosha, a notable mini-app of Taobao.

cs.IR↗

arXiv2025

WalletProbe: A Testing Framework for Browser-based Cryptocurrency Wallet Extensions

Serving as the first touch point for users to the cryptocurrency world, cryptocurrency wallets allow users to manage, receive, and transmit digital assets on blockchain networks and interact with emerging decentralized finance (DeFi) applications. Unfortunately, cryptocurrency wallets have always been the prime targets for attackers, and incidents of wallet breaches have been reported from time to time. Although some recent studies have characterized the vulnerabilities and scams related to wallets, they have generally been characterized in coarse granularity, overlooking potential risks inherent in detailed designs of cryptocurrency wallets, especially from perspectives including user interaction and advanced features. To fill the void, in this paper, we present a fine-grained security analysis on browser-based cryptocurrency wallets. To pinpoint security issues of components in wallets, we design WalletProbe, a mutation-based testing framework based on visual-level oracles. We have identified 13 attack vectors that can be abused by attackers to exploit cryptocurrency wallets and exposed 21 concrete attack strategies. By applying WalletProbe on 39 widely-adopted browser-based wallet extensions, we astonishingly figure out all of them can be abused to steal crypto assets from innocent users. Identified potential attack vectors were reported to wallet developers timely and 26 issues have been patched already. It is, hence, urgent for our community to take action to mitigate threats related to cryptocurrency wallets. We promise to release all code and data to promote the development of the community.

cs.CR↗

arXiv2025

Improve LLM-as-a-Judge Ability as a General Ability

LLM-as-a-Judge leverages the generative and reasoning capabilities of large language models (LLMs) to evaluate LLM responses across diverse scenarios, providing accurate preference signals. This approach plays a vital role in aligning LLMs with human values, ensuring ethical and reliable AI outputs that align with societal norms. Recent studies have raised many methods to train LLM as generative judges, but most of them are data consuming or lack accuracy, and only focus on LLM's judge ability. In this work, we regard judge ability as a general ability of LLM and implement a two-stage training approach, comprising supervised fine-tuning (SFT) warm-up and direct preference optimization (DPO) enhancement, to achieve judge style adaptation and improve judgment accuracy. Additionally, we introduce an efficient data synthesis method to generate judgmental content. Experimental results demonstrate that our approach, utilizing only about 2% to 40% of the data required by other methods, achieves SOTA performance on RewardBench. Furthermore, our training method enhances the general capabilities of the model by constructing complicated judge task, and the judge signals provided by our model have significantly enhanced the downstream DPO training performance of our internal models in our test to optimize policy model with Judge Model. We also open-source our model weights and training data to facilitate further research.

cs.CL↗

arXiv2025

Following Devils' Footprint: Towards Real-time Detection of Price Manipulation Attacks

Price manipulation attack is one of the notorious threats in decentralized finance (DeFi) applications, which allows attackers to exchange tokens at an extensively deviated price from the market. Existing efforts usually rely on reactive methods to identify such kind of attacks after they have happened, e.g., detecting attack transactions in the post-attack stage, which cannot mitigate or prevent price manipulation attacks timely. From the perspective of attackers, they usually need to deploy attack contracts in the pre-attack stage. Thus, if we can identify these attack contracts in a proactive manner, we can raise alarms and mitigate the threats. With the core idea in mind, in this work, we shift our attention from the victims to the attackers. Specifically, we propose SMARTCAT, a novel approach for identifying price manipulation attacks in the pre-attack stage proactively. For generality, it conducts analysis on bytecode and does not require any source code and transaction data. For accuracy, it depicts the control- and data-flow dependency relationships among function calls into a token flow graph. For scalability, it filters out those suspicious paths, in which it conducts inter-contract analysis as necessary. To this end, SMARTCAT can pinpoint attacks in real time once they have been deployed on a chain. The evaluation results illustrate that SMARTCAT significantly outperforms existing baselines with 91.6% recall and ~100% precision. Moreover, SMARTCAT also uncovers 616 attack contracts in-the-wild, accounting for \$9.25M financial losses, with only 19 cases publicly reported. By applying SMARTCAT as a real-time detector in Ethereum and Binance Smart Chain, it has raised 14 alarms 99 seconds after the corresponding deployment on average. These attacks have already led to $641K financial losses, and seven of them are still waiting for their ripe time.

cs.CR↗

arXiv2024

Yi-Lightning Technical Report

This technical report presents Yi-Lightning, our latest flagship large language model (LLM). It achieves exceptional performance, ranking 6th overall on Chatbot Arena, with particularly strong results (2nd to 4th place) in specialized categories including Chinese, Math, Coding, and Hard Prompts. Yi-Lightning leverages an enhanced Mixture-of-Experts (MoE) architecture, featuring advanced expert segmentation and routing mechanisms coupled with optimized KV-caching techniques. Our development process encompasses comprehensive pre-training, supervised fine-tuning (SFT), and reinforcement learning from human feedback (RLHF), where we devise deliberate strategies for multi-stage training, synthetic data construction, and reward modeling. Furthermore, we implement RAISE (Responsible AI Safety Engine), a four-component framework to address safety issues across pre-training, post-training, and serving phases. Empowered by our scalable super-computing infrastructure, all these innovations substantially reduce training, deployment and inference costs while maintaining high-performance standards. With further evaluations on public academic benchmarks, Yi-Lightning demonstrates competitive performance against top-tier LLMs, while we observe a notable disparity between traditional, static benchmark results and real-world, dynamic human preferences. This observation prompts a critical reassessment of conventional benchmarks' utility in guiding the development of more intelligent and powerful AI systems for practical applications. Yi-Lightning is now available through our developer platform at https://platform.lingyiwanwu.com.

cs.CL↗

arXiv2024

Remeasuring the Arbitrage and Sandwich Attacks of Maximal Extractable Value in Ethereum

Maximal Extractable Value (MEV) drives the prosperity of the blockchain ecosystem. By strategically including, excluding, or reordering transactions within blocks, block producers can extract additional value, which in turn incentivizes them to keep the decentralization of the whole blockchain platform. Before September 2022, around $675M was extracted in terms of MEV in Ethereum. Despite its importance, current work on identifying MEV activities suffers from two limitations. On the one hand, current methods heavily rely on clumsy heuristic rule-based patterns, leading to numerous false negatives or positives. On the other hand, the observations and conclusions are drawn from the early stage of Ethereum, which cannot be used as effective guiding principles after The Merge. To address these challenges, in this work, we innovatively proposed a profitability identification algorithm. Based on this, we designed two robust algorithms to identify MEV activities on our collected largest-ever dataset. Based on the identified results, we have characterized the overall landscape of the Ethereum MEV ecosystem, the impact the private transaction architectures bring in, and the adoption of back-running mechanisms. Our research sheds light on future MEV-related work.

cs.CR↗

arXiv2024

Yi: Open Foundation Models by 01.AI

We introduce the Yi model family, a series of language and multimodal models that demonstrate strong multi-dimensional capabilities. The Yi model family is based on 6B and 34B pretrained language models, then we extend them to chat models, 200K long context models, depth-upscaled models, and vision-language models. Our base models achieve strong performance on a wide range of benchmarks like MMLU, and our finetuned chat models deliver strong human preference rate on major evaluation platforms like AlpacaEval and Chatbot Arena. Building upon our scalable super-computing infrastructure and the classical transformer architecture, we attribute the performance of Yi models primarily to its data quality resulting from our data-engineering efforts. For pretraining, we construct 3.1 trillion tokens of English and Chinese corpora using a cascaded data deduplication and quality filtering pipeline. For finetuning, we polish a small scale (less than 10K) instruction dataset over multiple iterations such that every single instance has been verified directly by our machine learning engineers. For vision-language, we combine the chat language model with a vision transformer encoder and train the model to align visual representations to the semantic space of the language model. We further extend the context length to 200K through lightweight continual pretraining and demonstrate strong needle-in-a-haystack retrieval performance. We show that extending the depth of the pretrained checkpoint through continual pretraining further improves performance. We believe that given our current results, continuing to scale up model parameters using thoroughly optimized data will lead to even stronger frontier models.

cs.CL↗

arXiv2022

Disentangle and Remerge: Interventional Knowledge Distillation for Few-Shot Object Detection from A Conditional Causal Perspective

Few-shot learning models learn representations with limited human annotations, and such a learning paradigm demonstrates practicability in various tasks, e.g., image classification, object detection, etc. However, few-shot object detection methods suffer from an intrinsic defect that the limited training data makes the model cannot sufficiently explore semantic information. To tackle this, we introduce knowledge distillation to the few-shot object detection learning paradigm. We further run a motivating experiment, which demonstrates that in the process of knowledge distillation, the empirical error of the teacher model degenerates the prediction performance of the few-shot object detection model as the student. To understand the reasons behind this phenomenon, we revisit the learning paradigm of knowledge distillation on the few-shot object detection task from the causal theoretic standpoint, and accordingly, develop a Structural Causal Model. Following the theoretical guidance, we propose a backdoor adjustment-based knowledge distillation method for the few-shot object detection task, namely Disentangle and Remerge (D&R), to perform conditional causal intervention toward the corresponding Structural Causal Model. Empirically, the experiments on benchmarks demonstrate that D&R can yield significant performance boosts in few-shot object detection. Code is available at https://github.com/ZYN-1101/DandR.git.

cs.CV↗

arXiv2022

Supporting Medical Relation Extraction via Causality-Pruned Semantic Dependency Forest

Medical Relation Extraction (MRE) task aims to extract relations between entities in medical texts. Traditional relation extraction methods achieve impressive success by exploring the syntactic information, e.g., dependency tree. However, the quality of the 1-best dependency tree for medical texts produced by an out-of-domain parser is relatively limited so that the performance of medical relation extraction method may degenerate. To this end, we propose a method to jointly model semantic and syntactic information from medical texts based on causal explanation theory. We generate dependency forests consisting of the semantic-embedded 1-best dependency tree. Then, a task-specific causal explainer is adopted to prune the dependency forests, which are further fed into a designed graph convolutional network to learn the corresponding representation for downstream task. Empirically, the various comparisons on benchmark medical datasets demonstrate the effectiveness of our model.

cs.CL↗

arXiv2022

MME-CRS: Multi-Metric Evaluation Based on Correlation Re-Scaling for Evaluating Open-Domain Dialogue

Automatic open-domain dialogue evaluation is a crucial component of dialogue systems. Recently, learning-based evaluation metrics have achieved state-of-the-art performance in open-domain dialogue evaluation. However, these metrics, which only focus on a few qualities, are hard to evaluate dialogue comprehensively. Furthermore, these metrics lack an effective score composition approach for diverse evaluation qualities. To address the above problems, we propose a Multi-Metric Evaluation based on Correlation Re-Scaling (MME-CRS) for evaluating open-domain dialogue. Firstly, we build an evaluation metric composed of 5 groups of parallel sub-metrics called Multi-Metric Evaluation (MME) to evaluate the quality of dialogue comprehensively. Furthermore, we propose a novel score composition method called Correlation Re-Scaling (CRS) to model the relationship between sub-metrics and diverse qualities. Our approach MME-CRS ranks first on the final test data of DSTC10 track5 subtask1 Automatic Open-domain Dialogue Evaluation Challenge with a large margin, which proved the effectiveness of our proposed approach.

cs.CL↗

arXiv2021

Multiple Fusion Adaptation: A Strong Framework for Unsupervised Semantic Segmentation Adaptation

This paper challenges the cross-domain semantic segmentation task, aiming to improve the segmentation accuracy on the unlabeled target domain without incurring additional annotation. Using the pseudo-label-based unsupervised domain adaptation (UDA) pipeline, we propose a novel and effective Multiple Fusion Adaptation (MFA) method. MFA basically considers three parallel information fusion strategies, i.e., the cross-model fusion, temporal fusion and a novel online-offline pseudo label fusion. Specifically, the online-offline pseudo label fusion encourages the adaptive training to pay additional attention to difficult regions that are easily ignored by offline pseudo labels, therefore retaining more informative details. While the other two fusion strategies may look standard, MFA pays significant efforts to raise the efficiency and effectiveness for integration, and succeeds in injecting all the three strategies into a unified framework. Experiments on two widely used benchmarks, i.e., GTA5-to-Cityscapes and SYNTHIA-to-Cityscapes, show that our method significantly improves the semantic segmentation adaptation, and sets up new state of the art (58.2% and 62.5% mIoU, respectively). The code will be available at https://github.com/KaiiZhang/MFA.

cs.CV↗

arXiv2021

Cross Modification Attention Based Deliberation Model for Image Captioning

The conventional encoder-decoder framework for image captioning generally adopts a single-pass decoding process, which predicts the target descriptive sentence word by word in temporal order. Despite the great success of this framework, it still suffers from two serious disadvantages. Firstly, it is unable to correct the mistakes in the predicted words, which may mislead the subsequent prediction and result in error accumulation problem. Secondly, such a framework can only leverage the already generated words but not the possible future words, and thus lacks the ability of global planning on linguistic information. To overcome these limitations, we explore a universal two-pass decoding framework, where a single-pass decoding based model serving as the Drafting Model first generates a draft caption according to an input image, and a Deliberation Model then performs the polishing process to refine the draft caption to a better image description. Furthermore, inspired from the complementarity between different modalities, we propose a novel Cross Modification Attention (CMA) module to enhance the semantic expression of the image features and filter out error information from the draft captions. We integrate CMA with the decoder of our Deliberation Model and name it as Cross Modification Attention based Deliberation Model (CMA-DM). We train our proposed framework by jointly optimizing all trainable components from scratch with a trade-off coefficient. Experiments on MS COCO dataset demonstrate that our approach obtains significant improvements over single-pass decoding baselines and achieves competitive performances compared with other state-of-the-art two-pass decoding based methods.

cs.CV↗