arXiv subjects

Chao Ni

Publications and source records attributed to Chao Ni.

At least 19 recordsLinked to original sources

arXiv2026

PracRepair: LLM-Empowered Automated Program Repair Inspired by Human-Like Debugging Practices

As software systems grow in scale and complexity, debugging and repair remain costly and time-consuming. Large language models (LLMs) have advanced automated program repair (APR), but existing LLM-based APR approaches still largely rely on static or retrieved context, error messages, and coarse-grained validation outcomes. As a result, they underutilize dynamic information for failure understanding and repair, including failure-execution dynamics and patch-validation dynamics. Effectively leveraging such information, however, is challenging: failure-execution traces are large and noisy, raw static-dynamic context is not self-explanatory, and patch-validation dynamics are often reduced to coarse feedback. To address these challenges, we propose \textsc{PracRepair}, a fully automated LLM-based APR framework inspired by human-like debugging practices. \textsc{PracRepair} constructs an on-demand static-dynamic context from buggy programs and failure executions, performs question-driven failure diagnosis to formulate explicit repair hypotheses, and iteratively refines candidate patches using validation diagnostics and trace-level behavioral changes. Experimental results on Defects4J V1.2 and V2.0 show that \textsc{PracRepair} consistently outperforms state-of-the-art baselines. Specifically, under GPT-3.5, \textsc{PracRepair} correctly fixes 139/136 bugs on Defects4J V1.2/V2.0, while under GPT-4o it further improves to 162/171. Moreover, \textsc{PracRepair} generalizes effectively to RWB (Real-World Bugs), achieving the best performance across multiple foundation models.

cs.SE↗

arXiv2026

Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning

We describe the pre-training, post-training, and quantization of Nemotron 3 Super, a 120 billion (active 12 billion) parameter hybrid Mamba-Attention Mixture-of-Experts model. Nemotron 3 Super is the first model in the Nemotron 3 family to 1) be pre-trained in NVFP4, 2) leverage LatentMoE, a new Mixture-of-Experts architecture that optimizes for both accuracy per FLOP and accuracy per parameter, and 3) include MTP layers for inference acceleration through native speculative decoding. We pre-trained Nemotron 3 Super on 25 trillion tokens followed by post-training using supervised fine tuning (SFT) and reinforcement learning (RL). The final model supports up to 1M context length and achieves comparable accuracy on common benchmarks, while also achieving up to 2.2x and 7.5x higher inference throughput compared to GPT-OSS-120B and Qwen3.5-122B, respectively. Nemotron 3 Super datasets, along with the base, post-trained, and quantized checkpoints, are open-sourced on HuggingFace.

cs.LG↗

arXiv2026

SolAgent: A Specialized Multi-Agent Framework for Solidity Code Generation

Smart contracts are the backbone of the decentralized web, yet ensuring their functional correctness and security remains a critical challenge. While Large Language Models (LLMs) have shown promise in code generation, they often struggle with the rigorous requirements of smart contracts, frequently producing code that is buggy or vulnerable. To address this, we propose SolAgent, a novel tool-augmented multi-agent framework that mimics the workflow of human experts. SolAgent integrates a \textbf{dual-loop refinement mechanism}: an inner loop using the \textit{Forge} compiler to ensure functional correctness, and an outer loop leveraging the \textit{Slither} static analyzer to eliminate security vulnerabilities. Additionally, the agent is equipped with file system capabilities to resolve complex project dependencies. Experiments on the SolEval+ Benchmark, a rigorous suite derived from high-quality real-world projects, demonstrate that SolAgent achieves a Pass@1 rate of up to \textbf{64.39\%}, significantly outperforming state-of-the-art LLMs ($\sim$25\%), AI IDEs (e.g., GitHub Copilot), and existing agent frameworks. Moreover, it reduces security vulnerabilities by up to \textbf{39.77\%} compared to human-written baselines. Finally, we demonstrate that the high-quality trajectories generated by SolAgent can be used to distill smaller, open-source models, democratizing access to secure smart contract generation. We release our data and code at https://github.com/openpaperz/SolAgent.

cs.SE↗

arXiv2025

Pretraining Large Language Models with NVFP4

Large Language Models (LLMs) today are powerful problem solvers across many domains, and they continue to get stronger as they scale in model size, training set size, and training set quality, as shown by extensive research and experimentation across the industry. Training a frontier model today requires on the order of tens to hundreds of yottaflops, which is a massive investment of time, compute, and energy. Improving pretraining efficiency is therefore essential to enable the next generation of even more capable LLMs. While 8-bit floating point (FP8) training is now widely adopted, transitioning to even narrower precision, such as 4-bit floating point (FP4), could unlock additional improvements in computational speed and resource utilization. However, quantization at this level poses challenges to training stability, convergence, and implementation, notably for large-scale models trained on long token horizons. In this study, we introduce a novel approach for stable and accurate training of large language models (LLMs) using the NVFP4 format. Our method integrates Random Hadamard transforms (RHT) to bound block-level outliers, employs a two-dimensional quantization scheme for consistent representations across both the forward and backward passes, utilizes stochastic rounding for unbiased gradient estimation, and incorporates selective high-precision layers. We validate our approach by training a 12-billion-parameter model on 10 trillion tokens -- the longest publicly documented training run in 4-bit precision to date. Our results show that the model trained with our NVFP4-based pretraining technique achieves training loss and downstream task accuracies comparable to an FP8 baseline. These findings highlight that NVFP4, when combined with our training approach, represents a major step forward in narrow-precision LLM training algorithms.

cs.CL↗

arXiv2025

Navigating the Labyrinth: Path-Sensitive Unit Test Generation with Large Language Models

Unit testing is essential for software quality assurance, yet writing and maintaining tests remains time-consuming and error-prone. To address this challenge, researchers have proposed various techniques for automating unit test generation, including traditional heuristic-based methods and more recent approaches that leverage large language models (LLMs). However, these existing approaches are inherently path-insensitive because they rely on fixed heuristics or limited contextual information and fail to reason about deep control-flow structures. As a result, they often struggle to achieve adequate coverage, particularly for deep or complex execution paths. In this work, we present a path-sensitive framework, JUnitGenie, to fill this gap by combining code knowledge with the semantic capabilities of LLMs in guiding context-aware unit test generation. After extracting code knowledge from Java projects, JUnitGenie distills this knowledge into structured prompts to guide the generation of high-coverage unit tests. We evaluate JUnitGenie on 2,258 complex focal methods from ten real-world Java projects. The results show that JUnitGenie generates valid tests and improves branch and line coverage by 29.60% and 31.00% on average over both heuristic and LLM-based baselines. We further demonstrate that the generated test cases can uncover real-world bugs, which were later confirmed and fixed by developers.

cs.SE↗

arXiv2025

RepoTransAgent: Multi-Agent LLM Framework for Repository-Aware Code Translation

Repository-aware code translation is critical for modernizing legacy systems, enhancing maintainability, and enabling interoperability across diverse programming languages. While recent advances in large language models (LLMs) have improved code translation quality, existing approaches face significant challenges in practical scenarios: insufficient contextual understanding, inflexible prompt designs, and inadequate error correction mechanisms. These limitations severely hinder accurate and efficient translation of complex, real-world code repositories. To address these challenges, we propose RepoTransAgent, a novel multi-agent LLM framework for repository-aware code translation. RepoTransAgent systematically decomposes the translation process into specialized subtasks-context retrieval, dynamic prompt construction, and iterative code refinement-each handled by dedicated agents. Our approach leverages retrieval-augmented generation (RAG) for contextual information gathering, employs adaptive prompts tailored to varying repository scenarios, and introduces a reflection-based mechanism for systematic error correction. We evaluate RepoTransAgent on hundreds of Java-C# translation pairs from six popular open-source projects. Experimental results demonstrate that RepoTransAgent significantly outperforms state-of-the-art baselines in both compile and pass rates. Specifically, RepoTransAgent achieves up to 55.34% compile rate and 45.84% pass rate. Comprehensive analysis confirms the robustness and generalizability of RepoTransAgent across different LLMs, establishing its effectiveness for real-world repository-aware code translation.

cs.SE↗

arXiv2025

Enhancing Project-Specific Code Completion by Inferring Internal API Information

Project-specific code completion is a critical task that leverages context from a project to generate accurate code. State-of-the-art methods use retrieval-augmented generation (RAG) with large language models (LLMs) and project information for code completion. However, they often struggle to incorporate internal API information, which is crucial for accuracy, especially when APIs are not explicitly imported in the file. To address this, we propose a method to infer internal API information without relying on imports. Our method extends the representation of APIs by constructing usage examples and semantic descriptions, building a knowledge base for LLMs to generate relevant completions. We also introduce ProjBench, a benchmark that avoids leaked imports and consists of large-scale real-world projects. Experiments on ProjBench and CrossCodeEval show that our approach significantly outperforms existing methods, improving code exact match by 22.72% and identifier exact match by 18.31%. Additionally, integrating our method with existing baselines boosts code match by 47.80% and identifier match by 35.55%.

cs.SE↗

arXiv2025

Learning to Align Human Code Preferences

Large Language Models (LLMs) have demonstrated remarkable potential in automating software development tasks. While recent advances leverage Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) to align models with human preferences, the optimal training strategy remains unclear across diverse code preference scenarios. This paper systematically investigates the roles of SFT and DPO in aligning LLMs with different code preferences. Through both theoretical analysis and empirical observation, we hypothesize that SFT excels in scenarios with objectively verifiable optimal solutions, while applying SFT followed by DPO (S&D) enables models to explore superior solutions in scenarios without objectively verifiable optimal solutions. Based on the analysis and experimental evidence, we propose Adaptive Preference Optimization (APO), a dynamic integration approach that adaptively amplifies preferred responses, suppresses dispreferred ones, and encourages exploration of potentially superior solutions during training. Extensive experiments across six representative code preference tasks validate our theoretical hypotheses and demonstrate that APO consistently matches or surpasses the performance of existing SFT and S&D strategies. Our work provides both theoretical foundations and practical guidance for selecting appropriate training strategies in different code preference alignment scenarios.

cs.SE↗

arXiv2025

Detecting LLM-generated Code with Subtle Modification by Adversarial Training

With the rapid development of Large Language Models (LLMs), their powerful code-generation capabilities have been widely applied in tasks like code completion and automated development, demonstrating the value of improving coding efficiency. However, the extensive use of LLM-generated code also raises several new challenges. On the one hand, issues such as the regulation of code provenance, copyright disputes, and code quality have become increasingly concerning. How to effectively detect LLM-generated code and ensure its compliant and responsible use has become a critical and urgent issue. On the other hand, in practical applications, LLM-generated code is often subject to manual modifications, such as variable renaming or structural adjustments. Although some recent studies have proposed training-based and zero-shot methods for detecting LLM-generated code, these approaches show insufficient robustness when facing modified LLM-generated code, and there is a lack of an effective solution. To address the real-world scenario where LLM-generated code may undergo minor modifications, we propose CodeGPTSensor+, an enhanced version of CodeGPTSensor, which employs adversarial training to improve robustness against input perturbations. CodeGPTSensor+ integrates an adversarial sample generation module, Multi-objective Identifier and Structure Transformation (MIST), which systematically generates both high-quality and representative adversarial samples. This module effectively enhances the model's resistance against diverse adversarial attacks. Experimental results on the HMCorp dataset demonstrate that CodeGPTSensor+ significantly improves detection accuracy on the adversarial test set while maintaining high accuracy on the original test set, showcasing superior robustness compared to CodeGPTSensor.

cs.SE↗

arXiv2025

A Preference-Driven Methodology for High-Quality Solidity Code Generation

While Large Language Models (LLMs) have demonstrated remarkable progress in generating functionally correct Solidity code, they continue to face critical challenges in producing gas-efficient and secure code, which are critical requirements for real-world smart contract deployment. Although recent advances leverage Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) for code preference alignment, existing approaches treat functional correctness, gas optimization, and security as independent objectives, resulting in contracts that may achieve operational soundness but suffer from prohibitive execution costs or dangerous vulnerabilities. To address these limitations, we propose PrefGen, a novel framework that extends standard DPO beyond human preferences to incorporate quantifiable blockchain-specific metrics, enabling holistic multi-objective optimization specifically tailored for smart contract generation. Our framework introduces a comprehensive evaluation methodology with four complementary metrics: Pass@k (functional correctness), Compile@k (syntactic correctness), Gas@k (gas efficiency), and Secure@k (security assessment), providing rigorous multi-dimensional contract evaluation. Through extensive experimentation, we demonstrate that PrefGen significantly outperforms existing approaches across all critical dimensions, achieving 66.7% Pass@5, 58.9% Gas@5, and 62.5% Secure@5, while generating production-ready smart contracts that are functionally correct, cost-efficient, and secure.

cs.SE↗

arXiv2025

Enhancing LLM's Ability to Generate More Repository-Aware Unit Tests Through Precise Contextual Information Injection

Though many learning-based approaches have been proposed for unit test generation and achieved remarkable performance, they still have limitations in relying on task-specific datasets. Recently, Large Language Models (LLMs) guided by prompt engineering have gained attention for their ability to handle a broad range of tasks, including unit test generation. Despite their success, LLMs may exhibit hallucinations when generating unit tests for focal methods or functions due to their lack of awareness regarding the project's global context. These hallucinations may manifest as calls to non-existent methods, as well as incorrect parameters or return values, such as mismatched parameter types or numbers. While many studies have explored the role of context, they often extract fixed patterns of context for different models and focal methods, which may not be suitable for all generation processes (e.g., excessive irrelevant context could lead to redundancy, preventing the model from focusing on essential information). To overcome this limitation, we propose RATester, which enhances the LLM's ability to generate more repository-aware unit tests through global contextual information injection. To equip LLMs with global knowledge similar to that of human testers, we integrate the language server gopls, which provides essential features (e.g., definition lookup) to assist the LLM. When RATester encounters an unfamiliar identifier (e.g., an unfamiliar struct name), it first leverages gopls to fetch relevant definitions and documentation comments, and then uses this global knowledge to guide the LLM. By utilizing gopls, RATester enriches the LLM's knowledge of the project's global context, thereby reducing hallucinations during unit test generation.

cs.SE↗

arXiv2024

What You See Is What You Get: Attention-based Self-guided Automatic Unit Test Generation

Software defects heavily affect software's functionalities and may cause huge losses. Recently, many AI-based approaches have been proposed to detect defects, which can be divided into two categories: software defect prediction and automatic unit test generation. While these approaches have made great progress in software defect detection, they still have several limitations in practical application, including the low confidence of prediction models and the inefficiency of unit testing models. To address these limitations, we propose a WYSIWYG (i.e., What You See Is What You Get) approach: Attention-based Self-guided Automatic Unit Test GenERation (AUGER), which contains two stages: defect detection and error triggering. In the former stage, AUGER first detects the proneness of defects. Then, in the latter stage, it guides to generate unit tests for triggering such an error with the help of critical information obtained by the former stage. To evaluate the effectiveness of AUGER, we conduct a large-scale experiment by comparing with the state-of-the-art (SOTA) approaches on the widely used datasets (i.e., Bears, Bugs.jar, and Defects4J). AUGER makes great improvements by 4.7% to 35.3% and 17.7% to 40.4% in terms of F1-score and Precision in defect detection, and can trigger 23 to 84 more errors than SOTAs in unit test generation. Besides, we also conduct a further study to verify the generalization in practical usage by collecting a new dataset from real-world projects.

cs.SE↗

arXiv2024

Distinguishing LLM-generated from Human-written Code by Contrastive Learning

Large language models (LLMs), such as ChatGPT released by OpenAI, have attracted significant attention from both industry and academia due to their demonstrated ability to generate high-quality content for various tasks. Despite the impressive capabilities of LLMs, there are growing concerns regarding their potential risks in various fields, such as news, education, and software engineering. Recently, several commercial and open-source LLM-generated content detectors have been proposed, which, however, are primarily designed for detecting natural language content without considering the specific characteristics of program code. This paper aims to fill this gap by proposing a novel ChatGPT-generated code detector, CodeGPTSensor, based on a contrastive learning framework and a semantic encoder built with UniXcoder. To assess the effectiveness of CodeGPTSensor on differentiating ChatGPT-generated code from human-written code, we first curate a large-scale Human and Machine comparison Corpus (HMCorp), which includes 550K pairs of human-written and ChatGPT-generated code (i.e., 288K Python code pairs and 222K Java code pairs). Based on the HMCorp dataset, our qualitative and quantitative analysis of the characteristics of ChatGPT-generated code reveals the challenge and opportunity of distinguishing ChatGPT-generated code from human-written code with their representative features. Our experimental results indicate that CodeGPTSensor can effectively identify ChatGPT-generated code, outperforming all selected baselines.

cs.SE↗

arXiv2024

Improving the Ability of Pre-trained Language Model by Imparting Large Language Model's Experience

Large Language Models (LLMs) and pre-trained Language Models (LMs) have achieved impressive success on many software engineering tasks (e.g., code completion and code generation). By leveraging huge existing code corpora (e.g., GitHub), these models can understand the patterns in source code and use these patterns to predict code properties. However, LLMs under few-shot learning perform poorly on non-generative tasks (e.g., fault localization and vulnerability localization), and fine-tuning LLMs is time-consuming and costly for end users and small organizations. Furthermore, the performance of fine-tuning LMs for non-generative tasks is impressive, yet it heavily depends on the amount and quality of data. As a result, the current lack of data and the high cost of collecting it in real-world scenarios further limit the applicability of LMs. In this paper, we leverage the powerful generation capabilities of LLMs to enhance pre-trained LMs. Specifically, we use LLMs to generate domain-specific data, thereby improving the performance of pre-trained LMs on the target tasks. We conduct experiments by combining different LLMs in our generation phase and introducing various LMs to learn from the LLM-generated data. Then, we compare the performance of these LMs before and after learning the data. We find that LLM-generated data significantly enhances the performance of LMs. The improvement can reach up to 58.36% for fault localization and up to 6.09% for clone detection.

cs.SE↗

arXiv2024

Learning-based Models for Vulnerability Detection: An Extensive Study

Though many deep learning-based models have made great progress in vulnerability detection, we have no good understanding of these models, which limits the further advancement of model capability, understanding of the mechanism of model detection, and efficiency and safety of practical application of models. In this paper, we extensively and comprehensively investigate two types of state-of-the-art learning-based approaches (sequence-based and graph-based) by conducting experiments on a recently built large-scale dataset. We investigate seven research questions from five dimensions, namely model capabilities, model interpretation, model stability, ease of use of model, and model economy. We experimentally demonstrate the priority of sequence-based models and the limited abilities of both LLM (ChatGPT) and graph-based models. We explore the types of vulnerability that learning-based models skilled in and reveal the instability of the models though the input is subtlely semantical-equivalently changed. We empirically explain what the models have learned. We summarize the pre-processing as well as requirements for easily using the models. Finally, we initially induce the vital information for economically and safely practical usage of these models.

cs.SE↗

arXiv2024

ThinkRepair: Self-Directed Automated Program Repair

Though many approaches have been proposed for Automated Program Repair (APR) and indeed achieved remarkable performance, they still have limitations in fixing bugs that require analyzing and reasoning about the logic of the buggy program. Recently, large language models (LLMs) instructed by prompt engineering have attracted much attention for their powerful ability to address many kinds of tasks including bug-fixing. However, the quality of the prompt will highly affect the ability of LLMs and manually constructing high-quality prompts is a costly endeavor. To address this limitation, we propose a self-directed LLM-based automated program repair, ThinkRepair, with two main phases: collection phase and fixing phase. The former phase automatically collects various chains of thoughts that constitute pre-fixed knowledge by instructing LLMs with the Chain-of-Thought (CoT) prompt. The latter phase targets fixing a bug by first selecting examples for few-shot learning and second automatically interacting with LLMs, optionally appending with feedback of testing information. Evaluations on two widely studied datasets (Defects4J and QuixBugs) by comparing ThinkRepair with 12 SOTA APRs indicate the priority of ThinkRepair in fixing bugs. Notably, ThinkRepair fixes 98 bugs and improves baselines by 27%-344.4% on Defects4J V1.2. On Defects4J V2.0, ThinkRepair fixes 12-65 more bugs than the SOTA APRs. Additionally, ThinkRepair also makes a considerable improvement on QuixBugs (31 for Java and 21 for Python at most).

cs.SE↗

arXiv2024

Rectifier: Code Translation with Corrector via LLMs

Software migration is garnering increasing attention with the evolution of software and society. Early studies mainly relied on handcrafted translation rules to translate between two languages, the translation process is error-prone and time-consuming. In recent years, researchers have begun to explore the use of pre-trained large language models (LLMs) in code translation. However, code translation is a complex task that LLMs would generate mistakes during code translation, they all produce certain types of errors when performing code translation tasks, which include (1) compilation error, (2) runtime error, (3) functional error, and (4) non-terminating execution. We found that the root causes of these errors are very similar (e.g. failure to import packages, errors in loop boundaries, operator errors, and more). In this paper, we propose a general corrector, namely Rectifier, which is a micro and universal model for repairing translation errors. It learns from errors generated by existing LLMs and can be widely applied to correct errors generated by any LLM. The experimental results on translation tasks between C++, Java, and Python show that our model has effective repair ability, and cross experiments also demonstrate the robustness of our method.

cs.SE↗

arXiv2024

CasModaTest: A Cascaded and Model-agnostic Self-directed Framework for Unit Test Generation

Though many machine learning (ML)-based unit testing generation approaches have been proposed and indeed achieved remarkable performance, they still have several limitations in effectiveness and practical usage. More precisely, existing ML-based approaches (1) generate partial content of a unit test, mainly focusing on test oracle generation; (2) mismatch the test prefix with the test oracle semantically; and (3) are highly bound with the close-sourced model, eventually damaging data security. We propose CasModaTest, a cascaded, model-agnostic, and end-to-end unit test generation framework, to alleviate the above limitations with two cascaded stages: test prefix generation and test oracle generation. Then, we manually build large-scale demo pools to provide CasModaTest with high-quality test prefixes and test oracles examples. Finally, CasModaTest automatically assembles the generated test prefixes and test oracles and compiles or executes them to check their effectiveness, optionally appending with several attempts to fix the errors occurring in compiling and executing phases. To evaluate the effectiveness of CasModaTest, we conduct large-scale experiments on a widely used dataset (Defects4J) and compare it with four state-of-the-art (SOTA) approaches by considering two performance measures. The experimental results indicate that CasModaTest outperforms all SOTAs with a substantial improvement (i.e., 60.62%-352.55% in terms of accuracy, 2.83%-87.27% in terms of focal method coverage). Besides, we also conduct experiments of CasModaTest on different open-source LLMs and find that CasModaTest can also achieve significant improvements over SOTAs (39.82%-293.96% and 9.25%-98.95% in terms of accuracy and focal method coverage, respectively) in end-to-end unit test generation

cs.SE↗