Lingming Zhang

FAIR CodeGen team, Jade Copet, Quentin Carbonneaux et al. · meta-ai

We release Code World Model (CWM), a 32-billion-parameter open-weights LLM, to advance research on code generation with world models. To improve code understanding beyond what can be learned from training on static code alone, we mid-train CWM on a large amount of observation-action trajectories from Python interpreter and agentic Docker environments, and perform extensive multi-task reasoning RL in verifiable coding, math, and multi-turn software engineering environments. With CWM, we provide a strong testbed for researchers to explore the opportunities world modeling affords for improving code generation with reasoning and planning in computational environments. We present first steps of how world models can benefit agentic coding, enable step-by-step simulation of Python code execution, and show early results of how reasoning can benefit from the latter. CWM is a dense, decoder-only LLM trained with a context size of up to 131k tokens. Independent of its world modeling capabilities, CWM offers strong performance on general coding and math tasks: it reaches pass@1 scores of 65.8% on SWE-bench Verified (with test-time scaling), 68.6% on LiveCodeBench, 96.6% on Math-500, and 76.0% on AIME 2024. To support further research on code world modeling, we release model checkpoints after mid-training, SFT, and RL.

Chunqiu Steven Xia, Yinlin Deng, Soren Dunn et al.

Recent advancements in large language models (LLMs) have significantly advanced the automation of software development tasks, including code synthesis, program repair, and test generation. More recently, researchers and industry practitioners have developed various autonomous LLM agents to perform end-to-end software development tasks. These agents are equipped with the ability to use tools, run commands, observe feedback from the environment, and plan for future actions. However, the complexity of these agent-based approaches, together with the limited abilities of current LLMs, raises the following question: Do we really have to employ complex autonomous software agents? To attempt to answer this question, we build Agentless -- an agentless approach to automatically solve software development problems. Compared to the verbose and complex setup of agent-based approaches, Agentless employs a simplistic three-phase process of localization, repair, and patch validation, without letting the LLM decide future actions or operate with complex tools. Our results on the popular SWE-bench Lite benchmark show that surprisingly the simplistic Agentless is able to achieve both the highest performance (32.00%, 96 correct fixes) and low cost ($0.70) compared with all existing open-source software agents! Furthermore, we manually classified the problems in SWE-bench Lite and found problems with exact ground truth patch or insufficient/misleading issue descriptions. As such, we construct SWE-bench Lite-S by excluding such problematic issues to perform more rigorous evaluation and comparison. Our work highlights the current overlooked potential of a simple, interpretable technique in autonomous software development. We hope Agentless will help reset the baseline, starting point, and horizon for autonomous software agents, and inspire future work along this crucial direction.

Jiawei Liu, Jinjun Peng, Yuyao Wang et al.

Deep Learning (DL) is prevalently used in various industries to improve decision-making and automate processes, driven by the ever-evolving DL libraries and compilers. The correctness of DL systems is crucial for trust in DL applications. As such, the recent wave of research has been studying the automated synthesis of test-cases (i.e., DNN models and their inputs) for fuzzing DL systems. However, existing model generators only subsume a limited number of operators, lacking the ability to pervasively model operator constraints. To address this challenge, we propose NeuRI, a fully automated approach for generating valid and diverse DL models composed of hundreds of types of operators. NeuRI adopts a three-step process: (i) collecting valid and invalid API traces from various sources; (ii) applying inductive program synthesis over the traces to infer the constraints for constructing valid models; and (iii) using hybrid model generation which incorporates both symbolic and concrete operators. Our evaluation shows that NeuRI improves branch coverage of TensorFlow and PyTorch by 24% and 15% over the state-of-the-art model-level fuzzers. NeuRI finds 100 new bugs for PyTorch and TensorFlow in four months, with 81 already fixed or confirmed. Of these, 9 bugs are labelled as high priority or security vulnerability, constituting 10% of all high-priority bugs of the period. Open-source developers regard error-inducing tests reported by us as "high-quality" and "common in practice".

Junwei Liu, Kaixin Wang, Yixuan Chen et al.

The recent advance in Large Language Models (LLMs) has shaped a new paradigm of AI agents, i.e., LLM-based agents. Compared to standalone LLMs, LLM-based agents substantially extend the versatility and expertise of LLMs by enhancing LLMs with the capabilities of perceiving and utilizing external resources and tools. To date, LLM-based agents have been applied and shown remarkable effectiveness in Software Engineering (SE). The synergy between multiple agents and human interaction brings further promise in tackling complex real-world SE problems. In this work, we present a comprehensive and systematic survey on LLM-based agents for SE. We collect 124 papers and categorize them from two perspectives, i.e., the SE and agent perspectives. In addition, we discuss open challenges and future directions in this critical domain. The repository of this survey is at https://github.com/FudanSELab/Agent4SE-Paper-List.

Chunqiu Steven Xia, Lingming Zhang

Automated Program Repair (APR) aims to automatically generate patches for buggy programs. Recent APR work has been focused on leveraging modern Large Language Models (LLMs) to directly generate patches for APR. Such LLM-based APR tools work by first constructing an input prompt built using the original buggy code and then queries the LLM to generate patches. While the LLM-based APR tools are able to achieve state-of-the-art results, it still follows the classic Generate and Validate repair paradigm of first generating lots of patches and then validating each one afterwards. This not only leads to many repeated patches that are incorrect but also miss the crucial information in test failures as well as in plausible patches. To address these limitations, we propose ChatRepair, the first fully automated conversation-driven APR approach that interleaves patch generation with instant feedback to perform APR in a conversational style. ChatRepair first feeds the LLM with relevant test failure information to start with, and then learns from both failures and successes of earlier patching attempts of the same bug for more powerful APR. For earlier patches that failed to pass all tests, we combine the incorrect patches with their corresponding relevant test failure information to construct a new prompt for the LLM to generate the next patch. In this way, we can avoid making the same mistakes. For earlier patches that passed all the tests, we further ask the LLM to generate alternative variations of the original plausible patches. In this way, we can further build on and learn from earlier successes to generate more plausible patches to increase the chance of having correct patches. While our approach is general, we implement ChatRepair using state-of-the-art dialogue-based LLM -- ChatGPT. By calculating the cost of accessing ChatGPT, we can fix 162 out of 337 bugs for \$0.42 each!

Chunqiu Steven Xia, Matteo Paltenghi, Jia Le Tian et al.

Fuzzing has achieved tremendous success in discovering bugs and vulnerabilities in various software systems. Systems under test (SUTs) that take in programming or formal language as inputs, e.g., compilers, runtime engines, constraint solvers, and software libraries with accessible APIs, are especially important as they are fundamental building blocks of software development. However, existing fuzzers for such systems often target a specific language, and thus cannot be easily applied to other languages or even other versions of the same language. Moreover, the inputs generated by existing fuzzers are often limited to specific features of the input language, and thus can hardly reveal bugs related to other or new features. This paper presents Fuzz4All, the first fuzzer that is universal in the sense that it can target many different input languages and many different features of these languages. The key idea behind Fuzz4All is to leverage large language models (LLMs) as an input generation and mutation engine, which enables the approach to produce diverse and realistic inputs for any practically relevant language. To realize this potential, we present a novel autoprompting technique, which creates LLM prompts that are wellsuited for fuzzing, and a novel LLM-powered fuzzing loop, which iteratively updates the prompt to create new fuzzing inputs. We evaluate Fuzz4All on nine systems under test that take in six different languages (C, C++, Go, SMT2, Java and Python) as inputs. The evaluation shows, across all six languages, that universal fuzzing achieves higher coverage than existing, language-specific fuzzers. Furthermore, Fuzz4All has identified 98 bugs in widely used systems, such as GCC, Clang, Z3, CVC5, OpenJDK, and the Qiskit quantum computing platform, with 64 bugs already confirmed by developers as previously unknown.

Jiawei Liu, Jinkun Lin, Fabian Ruffy et al.

Deep-learning (DL) compilers such as TVM and TensorRT are increasingly being used to optimize deep neural network (DNN) models to meet performance, resource utilization and other requirements. Bugs in these compilers can result in models whose semantics differ from the original ones, producing incorrect results that corrupt the correctness of downstream applications. However, finding bugs in these compilers is challenging due to their complexity. In this work, we propose a new fuzz testing approach for finding bugs in deep-learning compilers. Our core approach consists of (i) generating diverse yet valid DNN test models that can exercise a large part of the compiler's transformation logic using light-weight operator specifications; (ii) performing gradient-based search to find model inputs that avoid any floating-point exceptional values during model execution, reducing the chance of missed bugs or false alarms; and (iii) using differential testing to identify bugs. We implemented this approach in NNSmith which has found 72 new bugs for TVM, TensorRT, ONNXRuntime, and PyTorch to date. Of these 58 have been confirmed and 51 have been fixed by their respective project maintainers.

Chunqiu Steven Xia, Lingming Zhang

Automated Program Repair (APR) can help developers automatically generate patches for bugs. Due to the impressive performance obtained using Large Pre-Trained Language Models (LLMs) on many code related tasks, researchers have started to directly use LLMs for APR. However, prior approaches simply repeatedly sample the LLM given the same constructed input/prompt created from the original buggy code, which not only leads to generating the same incorrect patches repeatedly but also miss the critical information in testcases. To address these limitations, we propose conversational APR, a new paradigm for program repair that alternates between patch generation and validation in a conversational manner. In conversational APR, we iteratively build the input to the model by combining previously generated patches with validation feedback. As such, we leverage the long-term context window of LLMs to not only avoid generating previously incorrect patches but also incorporate validation feedback to help the model understand the semantic meaning of the program under test. We evaluate 10 different LLM including the newly developed ChatGPT model to demonstrate the improvement of conversational APR over the prior LLM for APR approach.

Chunqiu Steven Xia, Yifeng Ding, Lingming Zhang

Automated Program Repair (APR) aspires to automatically generate patches for an input buggy program. Traditional APR tools typically focus on specific bug types and fixes through the use of templates, heuristics, and formal specifications. However, these techniques are limited in terms of the bug types and patch variety they can produce. As such, researchers have designed various learning-based APR tools with recent work focused on directly using Large Language Models (LLMs) for APR. While LLM-based APR tools are able to achieve state-of-the-art performance on many repair datasets, the LLMs used for direct repair are not fully aware of the project-specific information such as unique variable or method names. The plastic surgery hypothesis is a well-known insight for APR, which states that the code ingredients to fix the bug usually already exist within the same project. Traditional APR tools have largely leveraged the plastic surgery hypothesis by designing manual or heuristic-based approaches to exploit such existing code ingredients. However, as recent APR research starts focusing on LLM-based approaches, the plastic surgery hypothesis has been largely ignored. In this paper, we ask the following question: How useful is the plastic surgery hypothesis in the era of LLMs? Interestingly, LLM-based APR presents a unique opportunity to fully automate the plastic surgery hypothesis via fine-tuning and prompting. To this end, we propose FitRepair, which combines the direct usage of LLMs with two domain-specific fine-tuning strategies and one prompting strategy for more powerful APR. Our experiments on the widely studied Defects4j 1.2 and 2.0 datasets show that FitRepair fixes 89 and 44 bugs (substantially outperforming the best-performing baseline by 15 and 8), respectively, demonstrating a promising future of the plastic surgery hypothesis in the era of LLMs.

Chenyuan Yang, Yinlin Deng, Runyu Lu et al.

Compiler correctness is crucial, as miscompilation can falsify program behaviors, leading to serious consequences. Fuzzing has been studied to uncover compiler defects. However, compiler fuzzing remains challenging: Existing arts focus on black- and grey-box fuzzing, which generates tests without sufficient understanding of internal compiler behaviors. Meanwhile, traditional white-box techniques, like symbolic execution, are computationally inapplicable to the giant codebase of compilers. Recent advances demonstrate that Large Language Models (LLMs) excel in code generation/understanding tasks. Nonetheless, guiding LLMs with compiler source-code information remains a missing piece of research in compiler testing. To this end, we propose WhiteFox, the first white-box compiler fuzzer using LLMs with source-code information to test compiler optimization, with a spotlight on detecting deep logic bugs in the deep learning (DL) compilers. WhiteFox adopts a multi-agent framework: an LLM-based analysis agent examines the low-level optimization source code and produces requirements on the high-level test programs that can trigger the optimization; an LLM-based generation agent produces test programs based on the summarized requirements. Additionally, optimization-triggering tests are used as feedback to enhance the generation on the fly. Our evaluation on the three most popular DL compilers (i.e., PyTorch Inductor, TensorFlow-XLA, and TensorFlow Lite) shows WhiteFox can generate high-quality test programs to exercise deep optimizations, practicing up to 8X more than state-of-the-art fuzzers. WhiteFox has found 101 bugs for the DL compilers, with 92 confirmed as previously unknown and 70 fixed. WhiteFox has been acknowledged by the PyTorch team and is being incorporated into its development workflow. Beyond DL compilers, WhiteFox can also be adapted for compilers in different domains.

Jiawei Liu, Songrun Xie, Junhao Wang et al.

We introduce Differential Performance Evaluation (DPE), a framework designed to reliably evaluate Large Language Models (LLMs) for efficient code generation. Traditional coding benchmarks often fail to provide reliable insights into code efficiency, due to their reliance on simplistic test inputs and the absence of effective compound metrics. DPE addresses these issues by focusing on efficiency-demanding programming tasks and establishing an insightful compound metric for performance evaluation. DPE operates in two phases: To curate efficiency datasets, it selects efficiency-demanding tasks from existing coding benchmarks and generates computationally expensive inputs to stress the efficiency of LLM solutions. To assess the code efficiency, DPE profiles the new solution and compares it globally against a set of reference solutions that exhibit distinct efficiency levels, where the matched level defines its efficiency score. As a proof of concept, we use DPE to create EvalPerf, a benchmark with 121 performance-challenging coding tasks. Our comprehensive evaluation draws interesting findings on the efficiency impact of model sizes, instruction tuning, and prompting. For example, while the scaling law fails to account for code efficiency, general instruction tuning benefits both code correctness and efficiency. We also evaluate the evaluation by examining the effectiveness of DPE, showing that EvalPerf is reliable and convenient to use even across platforms.

Yuxiang Wei, Zhe Wang, Jiawei Liu et al.

We introduce Magicoder, a series of fully open-source (code, weights, and data) Large Language Models (LLMs) for code that significantly closes the gap with top code models while having no more than 7B parameters. Magicoder models are trained on 75K synthetic instruction data using OSS-Instruct, a novel approach to enlightening LLMs with open-source code snippets to generate diverse instruction data for code. Our main motivation is to mitigate the inherent bias of the synthetic data generated by LLMs through the wealth of open-source references for the production of more realistic and controllable data. The orthogonality of OSS-Instruct and other data generation methods like Evol-Instruct further enables us to build an enhanced MagicoderS. Both Magicoder and MagicoderS substantially outperform state-of-the-art code models with similar or even larger sizes on a wide range of coding benchmarks. Notably, MagicoderS-CL-7B based on CodeLlama even surpasses the prominent ChatGPT on HumanEval+ (66.5 vs. 65.9 in pass@1 ). Overall, OSS-Instruct opens a new direction for crafting diverse synthetic instruction data for code using abundant open-source references.

Yuxiang Wei, Olivier Duchenne, Jade Copet et al.

The recent DeepSeek-R1 release has demonstrated the immense potential of reinforcement learning (RL) in enhancing the general reasoning capabilities of large language models (LLMs). While DeepSeek-R1 and other follow-up work primarily focus on applying RL to competitive coding and math problems, this paper introduces SWE-RL, the first approach to scale RL-based LLM reasoning for real-world software engineering. Leveraging a lightweight rule-based reward (e.g., the similarity score between ground-truth and LLM-generated solutions), SWE-RL enables LLMs to autonomously recover a developer's reasoning processes and solutions by learning from extensive open-source software evolution data -- the record of a software's entire lifecycle, including its code snapshots, code changes, and events such as issues and pull requests. Trained on top of Llama 3, our resulting reasoning model, Llama3-SWE-RL-70B, achieves a 41.0% solve rate on SWE-bench Verified -- a human-verified collection of real-world GitHub issues. To our knowledge, this is the best performance reported for medium-sized (<100B) LLMs to date, even comparable to leading proprietary LLMs like GPT-4o. Surprisingly, despite performing RL solely on software evolution data, Llama3-SWE-RL has even emerged with generalized reasoning skills. For example, it shows improved results on five out-of-domain tasks, namely, function coding, library use, code reasoning, mathematics, and general language understanding, whereas a supervised-finetuning baseline even leads to performance degradation on average. Overall, SWE-RL opens up a new direction to improve the reasoning capabilities of LLMs through reinforcement learning on massive software engineering data.

Chunqiu Steven Xia, Yinlin Deng, Lingming Zhang

LLMs have become the go-to choice for code generation tasks, with an exponential increase in the training, development, and usage of LLMs specifically for code generation. To evaluate the ability of LLMs on code, both academic and industry practitioners rely on popular handcrafted benchmarks. However, prior benchmarks contain only a very limited set of problems, both in quantity and variety. Further, due to popularity and age, many benchmarks are prone to data leakage where example solutions can be readily found on the web and thus potentially in training data. Such limitations inevitably lead us to inquire: Is the leaderboard performance on existing benchmarks reliable and comprehensive enough to measure the program synthesis ability of LLMs? To address this, we introduce EvoEval -- a program synthesis benchmark suite created by evolving existing benchmarks into different targeted domains for a comprehensive evaluation of LLM coding abilities. Our study on 51 LLMs shows that compared to the high performance obtained on standard benchmarks like HumanEval, there is a significant drop in performance (on average 39.4%) when using EvoEval. Additionally, the decrease in performance can range from 19.6% to 47.7%, leading to drastic ranking changes amongst LLMs and showing potential overfitting of existing benchmarks. Furthermore, we showcase various insights, including the brittleness of instruction-following models when encountering rewording or subtle changes as well as the importance of learning problem composition and decomposition. EvoEval not only provides comprehensive benchmarks, but can be used to further evolve arbitrary problems to keep up with advances and the ever-changing landscape of LLMs for code. We have open-sourced our benchmarks, tools, and complete LLM generations at https://github.com/evo-eval/evoeval

Yifeng Ding, Jiawei Liu, Yuxiang Wei et al.

We introduce XFT, a simple yet powerful training scheme, by simply merging upcycled Mixture-of-Experts (MoE) to unleash the performance limit of instruction-tuned code Large Language Models (LLMs). While vanilla sparse upcycling fails to improve instruction tuning, XFT introduces a shared expert mechanism with a novel routing weight normalization strategy into sparse upcycling, which significantly boosts instruction tuning. After fine-tuning the upcycled MoE model, XFT introduces a learnable model merging mechanism to compile the upcycled MoE model back to a dense model, achieving upcycled MoE-level performance with only dense-model compute. By applying XFT to a 1.3B model, we create a new state-of-the-art tiny code LLM (<3B) with 67.1 and 64.6 pass@1 on HumanEval and HumanEval+ respectively. With the same data and model architecture, XFT improves supervised fine-tuning (SFT) by 13% on HumanEval+, along with consistent improvements from 2% to 13% on MBPP+, MultiPL-E, and DS-1000, demonstrating its generalizability. XFT is fully orthogonal to existing techniques such as Evol-Instruct and OSS-Instruct, opening a new dimension for improving code instruction tuning. Codes are available at https://github.com/ise-uiuc/xft.

Chunqiu Steven Xia, Zhe Wang, Yan Yang et al.

Large Language Models (LLMs) are reshaping almost all industries, including software engineering. In recent years, a number of LLM agents have been proposed to solve real-world software problems. Such software agents are typically equipped with a suite of coding tools and can autonomously decide the next actions to form complete trajectories to solve end-to-end software tasks. While promising, they typically require dedicated design and may still be suboptimal, since it can be extremely challenging and costly to exhaust the entire agent scaffold design space. Recognizing that software agents are inherently software themselves that can be further refined/modified, researchers have proposed a number of self-improving software agents recently, including the Darwin-Gödel Machine (DGM). Meanwhile, such self-improving agents require costly offline training on specific benchmarks and may not generalize well across different LLMs or benchmarks. In this paper, we propose Live-SWE-agent, the first live software agent that can autonomously and continuously evolve itself on-the-fly during runtime when solving real-world software problems. More specifically, Live-SWE-agent starts with the most basic agent scaffold with only access to bash tools (e.g., mini-SWE-agent), and autonomously evolves its own scaffold implementation while solving real-world software problems. Our evaluation on the widely studied SWE-bench Verified benchmark shows that Live-SWE-agent can achieve an impressive solve rate of 75.4% without test-time scaling, outperforming all existing open-source software agents and approaching the performance of the best proprietary solution. Moreover, Live-SWE-agent outperforms state-of-the-art manually crafted software agents on the recent SWE-Bench Pro benchmark, achieving the best-known solve rate of 45.8%.

Jiawei Liu, Chunqiu Steven Xia, Yuyao Wang et al.

Program synthesis has been long studied with recent approaches focused on directly using the power of Large Language Models (LLMs) to generate code. Programming benchmarks, with curated synthesis problems and test-cases, are used to measure the performance of various LLMs on code synthesis. However, these test-cases can be limited in both quantity and quality for fully assessing the functional correctness of the generated code. Such limitation in the existing benchmarks begs the following question: In the era of LLMs, is the code generated really correct? To answer this, we propose EvalPlus -- a code synthesis evaluation framework to rigorously benchmark the functional correctness of LLM-synthesized code. EvalPlus augments a given evaluation dataset with large amounts of test-cases newly produced by an automatic test input generator, powered by both LLM- and mutation-based strategies. While EvalPlus is general, we extend the test-cases of the popular HumanEval benchmark by 80x to build HumanEval+. Our extensive evaluation across 26 popular LLMs (e.g., GPT-4 and ChatGPT) demonstrates that HumanEval+ is able to catch significant amounts of previously undetected wrong code synthesized by LLMs, reducing the pass@k by up-to 19.3-28.9%. We also surprisingly found that test insufficiency can lead to mis-ranking. For example, both WizardCoder-CodeLlama and Phind-CodeLlama now outperform ChatGPT on HumanEval+, while none of them could on HumanEval. Our work not only indicates that prior popular code synthesis evaluation results do not accurately reflect the true performance of LLMs for code synthesis, but also opens up a new direction to improve such programming benchmarks through automated testing. We have open-sourced our tools, enhanced datasets as well as all LLM-generated code at https://github.com/evalplus/evalplus to facilitate and accelerate future LLM-for-code research.

Anjiang Wei, Yinlin Deng, Chenyuan Yang et al.

Deep learning (DL) systems can make our life much easier, and thus are gaining more and more attention from both academia and industry. Meanwhile, bugs in DL systems can be disastrous, and can even threaten human lives in safety-critical applications. To date, a huge body of research efforts have been dedicated to testing DL models. However, interestingly, there is still limited work for testing the underlying DL libraries, which are the foundation for building, optimizing, and running DL models. One potential reason is that test generation for the underlying DL libraries can be rather challenging since their public APIs are mainly exposed in Python, making it even hard to automatically determine the API input parameter types due to dynamic typing. In this paper, we propose FreeFuzz, the first approach to fuzzing DL libraries via mining from open source. More specifically, FreeFuzz obtains code/models from three different sources: 1) code snippets from the library documentation, 2) library developer tests, and 3) DL models in the wild. Then, FreeFuzz automatically runs all the collected code/models with instrumentation to trace the dynamic information for each covered API, including the types and values of each parameter during invocation, and shapes of input/output tensors. Lastly, FreeFuzz will leverage the traced dynamic information to perform fuzz testing for each covered API. The extensive study of FreeFuzz on PyTorch and TensorFlow, two of the most popular DL libraries, shows that FreeFuzz is able to automatically trace valid dynamic information for fuzzing 1158 popular APIs, 9X more than state-of-the-art LEMON with 3.5X lower overhead than LEMON. To date, FreeFuzz has detected 49 bugs for PyTorch and TensorFlow (with 38 already confirmed by developers as previously unknown).

Yuxiang Wei, Federico Cassano, Jiawei Liu et al.

Instruction tuning is a supervised fine-tuning approach that significantly improves the ability of large language models (LLMs) to follow human instructions. We propose SelfCodeAlign, the first fully transparent and permissive pipeline for self-aligning code LLMs without extensive human annotations or distillation. SelfCodeAlign employs the same base model for inference throughout the data generation process. It first extracts diverse coding concepts from high-quality seed snippets to generate new tasks. It then samples multiple responses per task, pairs each with test cases, and validates them in a sandbox environment. Finally, passing examples are selected for instruction tuning. In our primary experiments, we use SelfCodeAlign with CodeQwen1.5-7B to generate a dataset of 74k instruction-response pairs. Finetuning on this dataset leads to a model that achieves a 67.1 pass@1 on HumanEval+, surpassing CodeLlama-70B-Instruct despite being ten times smaller. Across all benchmarks, this finetuned model consistently outperforms the original version trained with OctoPack, the previous state-of-the-art method for instruction tuning without human annotations or distillation. Additionally, we show that SelfCodeAlign is effective across LLMs of various sizes, from 3B to 33B, and that the base models can benefit more from alignment with their own data distribution. We further validate each component's effectiveness in our pipeline, showing that SelfCodeAlign outperforms both direct distillation from GPT-4o and leading GPT-3.5-based distillation methods, such as OSS-Instruct and Evol-Instruct. SelfCodeAlign has also led to the creation of StarCoder2-Instruct, the first fully transparent, permissively licensed, and self-aligned code LLM that achieves state-of-the-art coding performance.

Chenyuan Yang, Zijie Zhao, Lingming Zhang

Bugs in operating system kernels can affect billions of devices and users all over the world. As a result, a large body of research has been focused on kernel fuzzing, i.e., automatically generating syscall (system call) sequences to detect potential kernel bugs or vulnerabilities. Kernel fuzzing aims to generate valid syscall sequences guided by syscall specifications that define both the syntax and semantics of syscalls. While there has been existing work trying to automate syscall specification generation, this remains largely manual work, and a large number of important syscalls are still uncovered. In this paper, we propose KernelGPT, the first approach to automatically synthesizing syscall specifications via Large Language Models (LLMs) for enhanced kernel fuzzing. Our key insight is that LLMs have seen massive kernel code, documentation, and use cases during pre-training, and thus can automatically distill the necessary information for making valid syscalls. More specifically, KernelGPT leverages an iterative approach to automatically infer the specifications, and further debug and repair them based on the validation feedback. Our results demonstrate that KernelGPT can generate more new and valid specifications and achieve higher coverage than state-of-the-art techniques. So far, by using newly generated specifications, KernelGPT has already detected 24 new unique bugs in Linux kernel, with 12 fixed and 11 assigned with CVE numbers. Moreover, a number of specifications generated by KernelGPT have already been merged into the kernel fuzzer Syzkaller, following the request from its development team.

Hwiwon Lee, Ziqi Zhang, Hanxiao Lu et al.

Rigorous security-focused evaluation of large language model (LLM) agents is imperative for establishing trust in their safe deployment throughout the software development lifecycle. However, existing benchmarks largely rely on synthetic challenges or simplified vulnerability datasets that fail to capture the complexity and ambiguity encountered by security engineers in practice. We introduce SEC-bench, the first fully automated benchmarking framework for evaluating LLM agents on authentic security engineering tasks. SEC-bench employs a novel multi-agent scaffold that automatically constructs code repositories with harnesses, reproduces vulnerabilities in isolated environments, and generates gold patches for reliable evaluation. Our framework automatically creates high-quality software vulnerability datasets with reproducible artifacts at a cost of only $0.87 per instance. Using SEC-bench, we implement two critical software security tasks to rigorously evaluate LLM agents' capabilities: proof-of-concept (PoC) generation and vulnerability patching. A comprehensive evaluation of state-of-the-art LLM code agents reveals significant performance gaps, achieving at most 18.0% success in PoC generation and 34.0% in vulnerability patching on our complete dataset. These results highlight the crucial steps needed toward developing LLM agents that are more practical, intelligent, and autonomous for security engineering.

Jiawei Liu, Nirav Diwan, Zhe Wang et al.

We introduce PurpCode, the first post-training recipe for training safe code reasoning models towards generating secure code and defending against malicious cyberactivities. PurpCode trains a reasoning model in two stages: (i) Rule Learning, which explicitly teaches the model to reference cybersafety rules to generate vulnerability-free code and to avoid facilitating malicious cyberactivities; and (ii) Reinforcement Learning, which optimizes model safety and preserves model utility through diverse, multi-objective reward mechanisms. To empower the training pipelines with comprehensive cybersafety data, we conduct internal red-teaming to synthesize comprehensive and high-coverage prompts based on real-world tasks for inducing unsafe cyberactivities in the model. Based on PurpCode, we develop a reasoning-based coding model, namely PurpCode-32B, which demonstrates state-of-the-art cybersafety, outperforming various frontier models. Meanwhile, our alignment method decreases the model overrefusal rates in both general and cybersafety-specific scenarios, while preserving model utility in both code generation and common security knowledge.

Jiawei Liu, Jia Le Tian, Vijay Daita et al.

Recent advances have been improving the context windows of Large Language Models (LLMs). To quantify the real long-context capabilities of LLMs, evaluators such as the popular Needle in a Haystack have been developed to test LLMs over a large chunk of raw texts. While effective, current evaluations overlook the insight of how LLMs work with long-context code, i.e., repositories. To this end, we initiate the RepoQA benchmark to evaluate LLMs on long-context code understanding. Traditional needle testers ask LLMs to directly retrieve the answer from the context without necessary deep understanding. In RepoQA, we built our initial task, namely Searching Needle Function (SNF), which exercises LLMs to search functions given their natural-language description, i.e., LLMs cannot find the desired function if they cannot understand the description and code. RepoQA is multilingual and comprehensive: it includes 500 code search tasks gathered from 50 popular repositories across 5 modern programming languages. By evaluating 26 general and code-specific LLMs on RepoQA, we show (i) there is still a small gap between the best open and proprietary models; (ii) different models are good at different languages; and (iii) models may understand code better without comments.

Yuxiang Wei, Chunqiu Steven Xia, Lingming Zhang

During Automated Program Repair (APR), it can be challenging to synthesize correct patches for real-world systems in general-purpose programming languages. Recent Large Language Models (LLMs) have been shown to be helpful "copilots" in assisting developers with various coding tasks, and have also been directly applied for patch synthesis. However, most LLMs treat programs as sequences of tokens, meaning that they are ignorant of the underlying semantics constraints of the target programming language. This results in plenty of statically invalid generated patches, impeding the practicality of the technique. Therefore, we propose Repilot, a general code generation framework to further copilot the AI "copilots" (i.e., LLMs) by synthesizing more valid patches during the repair process. Our key insight is that many LLMs produce outputs autoregressively (i.e., token by token), resembling human writing programs, which can be significantly boosted and guided through a Completion Engine. Repilot synergistically synthesizes a candidate patch through the interaction between an LLM and a Completion Engine, which 1) prunes away infeasible tokens suggested by the LLM and 2) proactively completes the token based on the suggestions provided by the Completion Engine. Our evaluation on a subset of the widely-used Defects4j 1.2 and 2.0 datasets shows that Repilot outperforms state-of-the-art techniques by fixing 27% and 47% more bugs, respectively. Moreover, Repilot produces more valid and correct patches than the base LLM with the same budget. While we focus on leveraging Repilot for APR in this work, the overall approach is also generalizable to other code generation tasks.

Jiawei Liu, Yuxiang Wei, Sen Yang et al.

In the past decade, Deep Learning (DL) systems have been widely deployed in various domains to facilitate our daily life. Meanwhile, it is extremely challenging to ensure the correctness of DL systems (e.g., due to their intrinsic nondeterminism), and bugs in DL systems can cause serious consequences and may even threaten human lives. In the literature, researchers have explored various techniques to test, analyze, and verify DL models, since their quality directly affects the corresponding system behaviors. Recently, researchers have also proposed novel techniques for testing the underlying operator-level DL libraries (such as TensorFlow and PyTorch), which provide general binary implementations for each high-level DL operator for running various DL models on many platforms. However, there is still limited work targeting the reliability of the emerging tensor compilers, which aim to directly compile high-level tensor computation graphs into high-performance binaries for better efficiency, portability, and scalability. In this paper, we target the important problem of tensor compiler testing, and have proposed Tzer, a practical fuzzing technique for the widely used TVM tensor compiler. Tzer focuses on mutating the low-level Intermediate Representation (IR) for TVM due to the limited mutation space for the high-level IR. More specifically, Tzer leverages both general-purpose and tensor-compiler-specific mutators guided by coverage feedback for evolutionary IR mutation; furthermore, Tzer also performs pass mutation in tandem with IR mutation for more effective fuzzing. Our results show that Tzer substantially outperforms existing fuzzing techniques on tensor compiler testing, with 75% higher coverage and 50% more valuable tests than the 2nd-best technique. To date, Tzer has detected 49 previously unknown bugs for TVM, with 37 bugs confirmed and 25 bugs fixed (PR merged).

Samuel Benton, Mengshi Zhang, Xia Li et al.

Program repair is an integral part of every software system's life-cycle but can be extremely challenging. To date, researchers have proposed various automated program repair (APR) techniques to reduce efforts of manual debugging. However, given a real-world buggy program, a typical APR technique usually generates a large number of patches, each of which needs to be validated against the original test suite which incurs extremely high computation costs. Although existing APR techniques have already leveraged various static and/or dynamic information to find the desired patches faster, they are still rather costly. In a recent work, researchers proposed unified debugging to leverage the patch execution information during APR to help boost fault localization; in this way,the application scope of APR techniques can be extended to all possible bugs, e.g., the patch execution information during APR can help with manual repair of the bugs that cannot be automatically fixed. Inspired by unified debugging, this work proposes SeAPR (Self-Boosted Automated Program Repair), the first technique to leverage the earlier patch execution information during APR to help boost automated repair itself on-the-fly. Our basic intuition is that patches similar to earlier high-quality/low-quality patches should be promoted/degraded to speed up the detection of the desired patches. This experimental study on 12 state-of-the-art APR systems demonstrates that, overall, SeAPR can substantially reduce the number of patch executions with negligible overhead. Our study also investigates the impact of various configurations on SeAPR. Lastly, our study demonstrates that SeAPR can even leverage the patch execution information from other APR tools from the same buggy program to further boost APR.

Lingchao Chen, Lingming Zhang

Generate-and-validate (G&V) automated program repair (APR) techniques have been extensively studied during the past decade. Meanwhile, such techniques can be extremely time-consuming due to manipulation of the program code to fabricate a large number of patches and also repeated executions of tests on patches to identify potential fixes. PraPR, a recent G&V APR technique, reduces these costs by modifying program code directly at the level of compiled bytecode, and further performing on-the-fly patching by allowing multiple patches to be tested within the same JVM session. However, PraPR is limited due to its pattern-based, bytecode-level nature and it is basically unsound/imprecise as it assumes that patch executions do not change global JVM state and affect later patch executions on the same JVM session. Inspired by the PraPR work, we propose a unified patch validation framework, named UniAPR, which aims to speed up the patch validation for both bytecode and source-code APR via on-the-fly patching; furthermore, UniAPR addresses the imprecise patch validation issue by resetting the JVM global state via runtime bytecode transformation. We have implemented UniAPR as a fully automated Maven Plugin. We have also performed the first study of on-the-fly patch validation for state-of-the-art source-code-level APR. Our experiments show the first empirical evidence that vanilla on-the-fly patch validation can be imprecise/unsound; in contrast, our UniAPR framework can speed up state-of-the-art APR by over an order of magnitude without incurring any imprecision in patch validation, enabling all existing APR techniques to explore a larger search space to fix more bugs in the near future. Furthermore, UniAPR directly enables hybrid source and bytecode APR to fix substantially more bugs than all state-of-the-art APR techniques (under the same time limit) in the near future.

Yiling Lou, Ali Ghanbari, Xia Li et al.

Software bugs are prevalent in modern software systems and notoriously hard to debug manually. Therefore, a large body of research efforts have been dedicated to automated software debugging, including both automated fault localization and program repair. However, the existing fault localization techniques are usually ineffective on real-world software systems while even the most advanced program repair techniques can only fix a small ratio of real-world bugs. Although fault localization and program repair are inherently connected, we observe that in the literature their only connection is that program repair techniques usually use off-the-shelf fault localization techniques (e.g., Ochiai) to determine the potential candidate statements/elements for patching. In this work, we explore their connection in the other direction, i.e., can program repair in turn help with fault localization? In this way,we not only open a new dimension for more powerful fault localization, but also extend the application scope of program repair to all possible bugs (not only the bugs that can be directly automatically fixed).We have designed ProFL, a simplistic approach using patch-execution results (from program repair) as the feedback information for fault localization. The experimental results on the widely used Defects4J benchmark show that the basic ProFL can already localize 161 of the 395 studied bugs within Top-1, while state-of-the-art spectrum and mutation based fault localization techniques at most localize 117 within Top-1. We also demonstrate ProFL's effectiveness under different settings. Lastly, we show that ProFL can further boost state-of-the-art fault localization via both unsupervised and supervised learning.

Haoran Wu, Xingya Wang, Jiehui Xu et al.

Smart contract is a special program that manages digital assets on blockchain. It is difficult to recover the loss if users make transactions through buggy smart contracts, which cannot be directly fixed. Hence, it is important to ensure the correctness of smart contracts before deploying them. This paper proposes a systematic framework to mutation testing for smart contracts on Ethereum, which is currently the most popular open blockchain for deploying and running smart contracts. Fifteen novel mutation operators have been designed for Ethereum Smart Contracts (ESC), in terms of keyword, global variable/function, variable unit, and error handling. An empirical study on 26 smart contracts in four Ethereum DApps has been conducted to evaluate the effectiveness of mutation testing. The experimental results show that our approach can outperform the coverage-based approach on defect detection rate (96.01% vs. 55.68%). The ESC mutation operators are effective to reveal real defects and we found 117 out of 729 real bug reports are related to our operators. These show the great potential of using mutation testing for quality assurance of ESC.

Husheng Zhou, Wei Li, Yuankun Zhu et al.

Deep Neural Networks (DNNs) have been widely applied in many autonomous systems such as autonomous driving. Recently, DNN testing has been intensively studied to automatically generate adversarial examples, which inject small-magnitude perturbations into inputs to test DNNs under extreme situations. While existing testing techniques prove to be effective, they mostly focus on generating digital adversarial perturbations (particularly for autonomous driving), e.g., changing image pixels, which may never happen in physical world. There is a critical missing piece in the literature on autonomous driving testing: understanding and exploiting both digital and physical adversarial perturbation generation for impacting steering decisions. In this paper, we present DeepBillboard, a systematic physical-world testing approach targeting at a common and practical driving scenario: drive-by billboards. DeepBillboard is capable of generating a robust and resilient printable adversarial billboard, which works under dynamic changing driving conditions including viewing angle, distance, and lighting. The objective is to maximize the possibility, degree, and duration of the steering-angle errors of an autonomous vehicle driving by the generated adversarial billboard. We have extensively evaluated the efficacy and robustness of DeepBillboard through conducting both digital and physical-world experiments. Results show that DeepBillboard is effective for various steering models and scenes. Furthermore, DeepBillboard is sufficiently robust and resilient for generating physical-world adversarial billboard tests for real-world driving under various weather conditions. To the best of our knowledge, this is the first study demonstrating the possibility of generating realistic and continuous physical-world tests for practical autonomous driving systems.

Ali Ghanbari, Lingming Zhang

Software debugging is tedious, time-consuming, and even error-prone by itself. So, various automated debugging techniques have been proposed in the literature to facilitate the debugging process. Automated Program Repair (APR) is one of the most recent advances in automated debugging, and can directly produce patches for buggy programs with minimal human intervention. Although various advanced APR techniques (including those that are either search-based or semantic-based) have been proposed, the simplistic mutation-based APR technique, which simply uses pre-defined mutation operators (e.g., changing a>=b into a>b) to mutate programs for finding patches, has not yet been thoroughly studied. In this paper, we implement the first practical bytecode-level APR technique, PraPR, and present the first extensive study on fixing real-world bugs (e.g., Defects4J bugs) using bytecode mutation. The experimental results show that surprisingly even PraPR with only the basic traditional mutators can produce genuine patches for 18 bugs. Furthermore, with our augmented mutators, PraPR is able to produce genuine patches for 43 bugs, significantly outperforming state-of-the-art APR. It is also an order of magnitude faster, indicating a promising future for bytecode-mutation-based APR.

Qi Luo, Kevin Moran, Lingming Zhang et al.

Test Case Prioritization (TCP) is an increasingly important regression testing technique for reordering test cases according to a pre-defined goal, particularly as agile practices gain adoption. To better understand these techniques, we perform the first extensive study aimed at empirically evaluating four static TCP techniques, comparing them with state-of-research dynamic TCP techniques across several quality metrics. This study was performed on 58 real-word Java programs encompassing 714 KLoC and results in several notable observations. First, our results across two effectiveness metrics (the Average Percentage of Faults Detected APFD and the cost cognizant APFDc) illustrate that at test-class granularity, these metrics tend to correlate, but this correlation does not hold at test-method granularity. Second, our analysis shows that static techniques can be surprisingly effective, particularly when measured by APFDc. Third, we found that TCP techniques tend to perform better on larger programs, but that program size does not affect comparative performance measures between techniques. Fourth, software evolution does not significantly impact comparative performance results between TCP techniques. Fifth, neither the number nor type of mutants utilized dramatically impact measures of TCP effectiveness under typical experimental settings. Finally, our similarity analysis illustrates that highly prioritized test cases tend to uncover dissimilar faults.

Mengshi Zhang, Yuqun Zhang, Lingming Zhang et al.

While Deep Neural Networks (DNNs) have established the fundamentals of DNN-based autonomous driving systems, they may exhibit erroneous behaviors and cause fatal accidents. To resolve the safety issues of autonomous driving systems, a recent set of testing techniques have been designed to automatically generate test cases, e.g., new input images transformed from the original ones. Unfortunately, many such generated input images often render inferior authenticity, lacking accurate semantic information of the driving scenes and hence compromising the resulting efficacy and reliability. In this paper, we propose DeepRoad, an unsupervised framework to automatically generate large amounts of accurate driving scenes to test the consistency of DNN-based autonomous driving systems across different scenes. In particular, DeepRoad delivers driving scenes with various weather conditions (including those with rather extreme conditions) by applying the Generative Adversarial Networks (GANs) along with the corresponding real-world weather scenes. Moreover, we have implemented DeepRoad to test three well-recognized DNN-based autonomous driving systems. Experimental results demonstrate that DeepRoad can detect thousands of behavioral inconsistencies in these systems.