We propose a new shared task for tactical data-to-text generation in the domain of source code libraries. Specifically, we focus on text generation of function descriptions from example software projects. Data is drawn from existing resources used for studying the related problem of semantic parser induction (Richardson and Kuhn, 2017b; Richardson and Kuhn, 2017a), and spans a wide variety of both natural languages and programming languages. In this paper, we describe these existing resources, which will serve as training and development data for the task, and discuss plans for building new independent test sets.
Representing code changes as numeric feature vectors, i.e., code change representations, is usually an essential step to automate many software engineering tasks related to code changes, e.g., commit message generation and just-in-time defect prediction. Intuitively, the quality of code change representations is crucial for the effectiveness of automated approaches. Prior work on code changes usually designs and evaluates code change representation approaches for a specific task, and little work has investigated code change encoders that can be used and jointly trained on various tasks. To fill this gap, this work proposes a novel Code Change Representation learning approach named CCRep, which can learn to encode code changes as feature vectors for diverse downstream tasks. Specifically, CCRep regards a code change as the combination of its before-change and after-change code, leverages a pre-trained code model to obtain high-quality contextual embeddings of code, and uses a novel mechanism named query back to extract and encode the changed code fragments and make them explicitly interact with the whole code change. To evaluate CCRep and demonstrate its applicability to diverse code-change-related tasks, we apply it to three tasks: commit message generation, patch correctness assessment, and just-in-time defect prediction. Experimental results show that CCRep outperforms the state-of-the-art techniques on each task.
We introduce OSVBench, a new benchmark for evaluating Large Language Models (LLMs) in generating complete specification code pertaining to operating system kernel verification tasks. The benchmark first defines the specification generation problem into a program synthesis problem within a confined scope of syntax and semantics by providing LLMs with the programming model. The LLMs are required to understand the provided verification assumption and the potential syntax and semantics space to search for, then generate the complete specification for the potentially buggy operating system code implementation under the guidance of the high-level functional description of the operating system. This benchmark is built upon a real-world operating system kernel, Hyperkernel, and consists of 245 complex specification generation tasks in total, each is a long context task of about 20k-30k tokens. Our comprehensive evaluation of 12 LLMs exhibits the limited performance of the current LLMs on the specification generation tasks for operating system verification. Significant disparities in their performance on the benchmark highlight differences in their ability to handle long-context code generation tasks. The evaluation toolkit and benchmark are available at https://github.com/lishangyu-hkust/OSVBench.
Instruction tuning -- tuning large language models on instruction-output pairs -- is a promising technique for making models better adapted to the real world. Yet, the key factors driving the model's capability to understand and follow instructions not seen during training remain under-explored. Our investigation begins with a series of synthetic experiments within the theoretical framework of a Turing-complete algorithm called Markov algorithm, which allows fine-grained control over the instruction-tuning data. Generalization and robustness with respect to the training distribution emerge once a diverse enough set of tasks is provided, even though very few examples are provided for each task. We extend these initial results to a real-world application scenario of code generation and find that a more diverse instruction set, extending beyond code-related tasks, improves the performance of code generation. Our observations suggest that a more diverse semantic space for instruction-tuning sets greatly improves the model's ability to follow instructions and perform tasks.
Recent advances in agentic LLMs have demonstrated great capabilities in Verilog code generation. However, existing approaches either use LLM-assisted single-agent prompting or cooperation-only multi-agent learning, which will lead to: (i) Degeneration issue for single-agent learning: characterized by diminished error detection and correction capabilities; (ii) Error propagation in cooperation-only multi-agent learning: erroneous information from the former agent will be propagated to the latter through prompts, which can make the latter agents generate buggy code. In this paper, we propose an LLM-based coopetitive multi-agent prompting framework, in which the agents cannot collaborate with each other to form the generation pipeline, but also create a healthy competitive mechanism to improve the generating quality. Our experimental results show that the coopetitive multi-agent framework can effectively mitigate the degeneration risk and reduce the error propagation while improving code error correction capabilities, resulting in higher quality Verilog code generation. The effectiveness of our approach is validated through extensive experiments. On VerilogEval Machine and Human dataset, CoopetitiveV+GPT-4 achieves 99.2% and 99.1% pass@10 scores, respectively. While on RTLLM, CoopetitiveV+GPT-4 obtains 100% syntax and 99.9% functionality pass@5 scores.
In recent times, it has been shown that one can use code as data to aid various applications such as automatic commit message generation, automatic generation of pull request descriptions and automatic program repair. Take for instance the problem of commit message generation. Treating source code as a sequence of tokens, state of the art techniques generate commit messages using neural machine translation models. However, they tend to ignore the syntactic structure of programming languages. Previous work, i.e., code2seq has used structural information from Abstract Syntax Tree (AST) to represent source code and they use it to automatically generate method names. In this paper, we elaborate upon this state of the art approach and modify it to represent source code edits. We determine the effect of using such syntactic structure for the problem of classifying code edits. Inspired by the code2seq approach, we evaluate how using structural information from AST, i.e., paths between AST leaf nodes can help with the task of code edit classification on two datasets of fine-grained syntactic edits. Our experiments shows that attempts of adding syntactic structure does not result in any improvements over less sophisticated methods. The results suggest that techniques such as code2seq, while promising, have a long way to go before they can be generically applied to learning code edit representations. We hope that these results will benefit other researchers and inspire them to work further on this problem.
We study the code generation behavior of instruction-tuned models built on top of code pre-trained language models when they could access an auxiliary function to implement a function. We design several ways to provide auxiliary functions to the models by adding them to the query or providing a response prefix to incorporate the ability to utilize auxiliary functions with the instruction-following capability. Our experimental results show the effectiveness of combining the base models' auxiliary function utilization ability with the instruction following ability. In particular, the performance of adopting our approaches with the open-sourced language models surpasses that of the recent powerful proprietary language models, i.e., gpt-4o.
The design flow of processors, particularly in hardware description languages (HDL) like Verilog and Chisel, is complex and costly. While recent advances in large language models (LLMs) have significantly improved coding tasks in software languages such as Python, their application in HDL generation remains limited due to the scarcity of high-quality HDL data. Traditional methods of adapting LLMs for hardware design rely on synthetic HDL datasets, which often suffer from low quality because even advanced LLMs like GPT perform poorly in the HDL domain. Moreover, these methods focus solely on chat tasks and the Verilog language, limiting their application scenarios. In this paper, we observe that: (1) HDL code collected from the real world is of higher quality than code generated by LLMs. (2) LLMs like GPT-3.5 excel in summarizing HDL code rather than generating it. (3) An explicit language tag can help LLMs better adapt to the target language when there is insufficient data. Based on these observations, we propose an efficient LLM fine-tuning pipeline for HDL generation that integrates a multi-level summarization data synthesis process with a novel Chat-FIM-Tag supervised fine-tuning method. The pipeline enhances the generation of HDL code from natural language descriptions and enables the handling of various tasks such as chat and infilling incomplete code. Utilizing this pipeline, we introduce CodeV, a series of HDL generation LLMs. Among them, CodeV-All not only possesses a more diverse range of language abilities, i.e. Verilog and Chisel, and a broader scope of tasks, i.e. Chat and fill-in-middle (FIM), but it also achieves performance on VerilogEval that is comparable to or even surpasses that of CodeV-Verilog fine-tuned on Verilog only, making them the first series of open-source LLMs designed for multi-scenario HDL generation.
Large language models of code (Code-LLMs) have recently brought tremendous advances to code completion, a fundamental feature of programming assistance and code intelligence. However, most existing works ignore the possible presence of bugs in the code context for generation, which are inevitable in software development. Therefore, we introduce and study the buggy-code completion problem, inspired by the realistic scenario of real-time code suggestion where the code context contains potential bugs -- anti-patterns that can become bugs in the completed program. To systematically study the task, we introduce two datasets: one with synthetic bugs derived from semantics-altering operator changes (buggy-HumanEval) and one with realistic bugs derived from user submissions to coding problems (buggy-FixEval). We find that the presence of potential bugs significantly degrades the generation performance of the high-performing Code-LLMs. For instance, the passing rates of CODEGEN-2B-MONO on test cases of buggy-HumanEval drop more than 50% given a single potential bug in the context. Finally, we investigate several post-hoc methods for mitigating the adverse effect of potential bugs and find that there remains a significant gap in post-mitigation performance.
Large language models (LLMs) have achieved exceptional performance in code generation. However, the performance remains unsatisfactory in generating library-oriented code, especially for the libraries not present in the training data of LLMs. Previous work utilizes API recommendation technology to help LLMs use libraries: it retrieves APIs related to the user requirements, then leverages them as context to prompt LLMs. However, developmental requirements can be coarse-grained, requiring a combination of multiple fine-grained APIs. This granularity inconsistency makes API recommendation a challenging task. To address this, we propose CAPIR (Compositional API Recommendation), which adopts a "divide-and-conquer" strategy to recommend APIs for coarse-grained requirements. Specifically, CAPIR employs an LLM-based Decomposer to break down a coarse-grained task description into several detailed subtasks. Then, CAPIR applies an embedding-based Retriever to identify relevant APIs corresponding to each subtask. Moreover, CAPIR leverages an LLM-based Reranker to filter out redundant APIs and provides the final recommendation. To facilitate the evaluation of API recommendation methods on coarse-grained requirements, we present two challenging benchmarks, RAPID (Recommend APIs based on Documentation) and LOCG (Library-Oriented Code Generation). Experimental results on these benchmarks, demonstrate the effectiveness of CAPIR in comparison to existing baselines. Specifically, on RAPID's Torchdata-AR dataset, compared to the state-of-the-art API recommendation approach, CAPIR improves recall@5 from 18.7% to 43.2% and precision@5 from 15.5% to 37.1%. On LOCG's Torchdata-Code dataset, compared to code generation without API recommendation, CAPIR improves pass@100 from 16.0% to 28.0%.
As the adoption of LLMs becomes more widespread in software coding ecosystems, a pressing issue has emerged: does the generated code contain social bias and unfairness, such as those related to age, gender, and race? This issue concerns the integrity, fairness, and ethical foundation of software applications that depend on the code generated by these models but are underexplored in the literature. This paper presents a novel bias testing framework that is specifically designed for code generation tasks. Based on this framework, we conduct an extensive empirical study on the biases in code generated by five widely studied LLMs (i.e., PALM-2-CodeChat-bison, Claude-instant-1, GPT-3.5-turbo, GPT-4-turbo, and GPT-4). Our findings reveal that biases are prevalent. For example, 13.47% to 49.10% of the codes generated by these LLMs have biased behaviors towards gender. Moreover, we study five bias mitigation prompt strategies that are commonly used in current code generation scenarios, i.e., zero-shot, one-shot, few-shot, and two Chain-of-Thought (CoT) prompts, with and without provided feedback-driven refinement. Our evaluation results illustrate that using direct prompt engineering strategies has limited effectiveness in mitigating bias, but our test execution feedback can help to reduce the ratio of code biases to a large extent (e.g., from 59.88% to 4.79% for GPT-4).
Large Language Models (LLMs) are widely used for tasks such as natural language and code generation, but their outputs often suffer from issues like hallucination, toxicity, and incorrect results. Current libraries for structured LLM generation rely on left-to-right decoding without support for backtracking, limiting the ability to correct or refine outputs mid-generation. To address this, we introduce IterGen, a user-friendly library for iterative, grammar-guided LLM generation that enables users to move both forward and backward within the generated output based on grammar symbols. By leveraging a symbol-to-position mapping and maintaining the key-value (KV) cache state, IterGen ensures efficient and structured generation while allowing for corrections during the process. We demonstrate IterGen's effectiveness in two important applications: reducing privacy leakage in LLM outputs and improving the accuracy of LLM-generated SQL and Vega-Lite queries. Our code and additional resources are available at https://structuredllm.com.
Pre-trained Generative Language models (e.g. PLBART, CodeT5, SPT-Code) for source code yielded strong results on several tasks in the past few years, including code generation and translation. These models have adopted varying pre-training objectives to learn statistics of code construction from very large-scale corpora in a self-supervised fashion; the success of pre-trained models largely hinges on these pre-training objectives. This paper proposes a new pre-training objective, "Naturalizing" of source code, exploiting code's bimodal, dual-channel (formal & natural channels) nature. Unlike natural language, code's bimodal, dual-channel nature allows us to generate semantically equivalent code at scale. We introduce six classes of semantic preserving transformations to introduce un-natural forms of code, and then force our model to produce more natural original programs written by developers. Learning to generate equivalent, but more natural code, at scale, over large corpora of open-source code, without explicit manual supervision, helps the model learn to both ingest & generate code. We fine-tune our model in three generative Software Engineering tasks: code generation, code translation, and code refinement with limited human-curated labeled data and achieve state-of-the-art performance rivaling CodeT5. We show that our pre-trained model is especially competitive at zero-shot and few-shot learning, and better at learning code properties (e.g., syntax, data flow).
In this work, we tackle the problem of structured text generation, specifically academic paper generation in $\LaTeX{}$, inspired by the surprisingly good results of basic character-level language models. Our motivation is using more recent and advanced methods of language modeling on a more complex dataset of $\LaTeX{}$ source files to generate realistic academic papers. Our first contribution is preparing a dataset with $\LaTeX{}$ source files on recent open-source computer vision papers. Our second contribution is experimenting with recent methods of language modeling and text generation such as Transformer and Transformer-XL to generate consistent $\LaTeX{}$ code. We report cross-entropy and bits-per-character (BPC) results of the trained models, and we also discuss interesting points on some examples of the generated $\LaTeX{}$ code.
Since the rise of neural natural-language-to-code models (NL->Code) that can generate long expressions and statements rather than a single next-token, one of the major problems has been reliably evaluating their generated output. In this paper, we propose CodeBERTScore: an evaluation metric for code generation, which builds on BERTScore (Zhang et al., 2020). Instead of encoding only the generated tokens as in BERTScore, CodeBERTScore also encodes the natural language input preceding the generated code, thus modeling the consistency between the generated code and its given natural language context as well. We perform an extensive evaluation of CodeBERTScore across four programming languages. We find that CodeBERTScore achieves a higher correlation with human preference and with functional correctness than all existing metrics. That is, generated code that receives a higher score by CodeBERTScore is more likely to be preferred by humans, as well as to function correctly when executed. We release five language-specific pretrained models to use with our publicly available code. Our language-specific models have been downloaded more than 1,000,000 times from the Huggingface Hub. Our code and data are available at https://github.com/neulab/code-bert-score
With the rapid advancement of large language models (LLMs), extensive research has been conducted to investigate the code generation capabilities of LLMs. However, existing efforts primarily focus on general-domain tasks, leaving LLMs' code generation performance in real-world application domains underexplored. This raises a critical question: can a model's general-domain coding ability reliably represent its ability in specialized domains? In this paper, we introduce DomainCodeBench, a multi-domain code generation benchmark designed to systematically evaluate LLMs across 12 software application domains and 15 programming languages. DomainCodeBench contains 2,400 manually verified tasks with ground truth, human-annotated docstrings, and fine-grained dependency information to ensure more coverage of domain-specific challenges. Specifically, we first identify the most popular application domains by topic mining. Then, we curate coding tasks based on commonly used frameworks and platforms in each domain. We obtain several findings through extensive experiments on DomainCodeBench with ten mainstream LLMs. (1) Performance decoupling: experiments reveal that top general-domain models do not consistently excel in specific application domains; (2) Domain-specific weaknesses: LLMs often fail due to domain knowledge gaps and third-party library misusage; (3) Contextual enhancement: we show that augmenting prompts with domain-specific knowledge improves performance by around 38.17%, providing actionable insights for performance optimization. Our replication package, including the benchmark, source code, and experimental results, is available at https://github.com/DeepSoftwareAnalytics/DomainCodeBench.
Reimplementing solutions to previously solved software engineering problems is not only inefficient but also introduces inadequate and error-prone code. Many existing methods achieve impressive performance on this issue by using autoregressive text-generation models trained on code. However, these methods are not without their flaws. The generated code from these models can be buggy, lack documentation, and introduce vulnerabilities that may go unnoticed by developers. An alternative to code generation -- neural code search -- is a field of machine learning where a model takes natural language queries as input and, in turn, relevant code samples from a database are returned. Due to the nature of this pre-existing database, code samples can be documented, tested, licensed, and checked for vulnerabilities before being used by developers in production. In this work, we present CodeDSI, an end-to-end unified approach to code search. CodeDSI is trained to directly map natural language queries to their respective code samples, which can be retrieved later. In an effort to improve the performance of code search, we have investigated docid representation strategies, impact of tokenization on docid structure, and dataset sizes on overall code search performance. Our results demonstrate CodeDSI strong performance, exceeding conventional robust baselines by 2-6% across varying dataset sizes.
We study the problem of generating source code in a strongly typed, Java-like programming language, given a label (for example a set of API calls or types) carrying a small amount of information about the code that is desired. The generated programs are expected to respect a "realistic" relationship between programs and labels, as exemplified by a corpus of labeled programs available during training. Two challenges in such conditional program generation are that the generated programs must satisfy a rich set of syntactic and semantic constraints, and that source code contains many low-level features that impede learning. We address these problems by training a neural generator not on code but on program sketches, or models of program syntax that abstract out names and operations that do not generalize across programs. During generation, we infer a posterior distribution over sketches, then concretize samples from this distribution into type-safe programs using combinatorial techniques. We implement our ideas in a system for generating API-heavy Java code, and show that it can often predict the entire body of a method given just a few API calls or data types that appear in the method.
Large language models (LLMs) have demonstrated an impressive ability to generate codes on competitive programming tasks. However, with limited sample numbers, LLMs still suffer from poor accuracy. Inspired by the process of human programming, we propose a generate-and-edit approach named Self-Edit that utilizes execution results of the generated code from LLMs to improve the code quality on the competitive programming task. We execute the generated code on the example test case provided in the question and wrap execution results into a supplementary comment. Utilizing this comment as guidance, our fault-aware code editor is employed to correct errors in the generated code. We perform extensive evaluations across two competitive programming datasets with nine different LLMs. Compared to directly generating from LLMs, our approach can improve the average of pass@1 by 89\% on APPS-dev, 31\% on APPS-test, and 48\% on HumanEval over nine popular code generation LLMs with parameter sizes ranging from 110M to 175B. Compared to other post-processing methods, our method demonstrates superior accuracy and efficiency.
We investigate how provision of in-context examples based on discourse similarity affects quality and hallucination in diagram code generation from educational materials. Using a sample of texts as sources for diagram generation, we test two methods using a conventional text-to-diagram task format, instruction-based zero-shot generation and ICL with text-to-diagram examples, and an unconventional ICL method with a discourse analysis-to-diagram example, which requires a model to generate a diagram from a discourse analysis. We evaluate AI-generated diagrams for faithfulness to provided context and their quality outcomes using both human and automated evaluation. Bayesian data analysis of the human diagram evaluation shows that there is significant evidence that the unconventional task format improves context faithfulness in diagram generation compared to the zero-shot instruction with reducing the odds of context faithfulness hallucination by nearly $70\%$; however, we do not find the ICL having a significant effect on the diagram aesthetic presentation and superficial coherence. Evaluation with the LLM-as-a-judge G-Eval metric shows nearly 4.5-fold increased odds of the perfect faithfulness outcome for the novel ICL task. Our study provides a novel insight into ICL format effect on superficial fluency and faithfulness in context-grounded structured diagram code generation.