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.
Humans write code in a fundamentally interactive manner and rely on constant execution feedback to correct errors, resolve ambiguities, and decompose tasks. While LLMs have recently exhibited promising coding capabilities, current coding benchmarks mostly consider a static instruction-to-code sequence transduction process, which has the potential for error propagation and a disconnect between the generated code and its final execution environment. To address this gap, we introduce InterCode, a lightweight, flexible, and easy-to-use framework of interactive coding as a standard reinforcement learning (RL) environment, with code as actions and execution feedback as observations. Our framework is language and platform agnostic, uses self-contained Docker environments to provide safe and reproducible execution, and is compatible out-of-the-box with traditional seq2seq coding methods, while enabling the development of new methods for interactive code generation. We use InterCode to create three interactive code environments with Bash, SQL, and Python as action spaces, leveraging data from the static NL2Bash, Spider, and MBPP datasets. We demonstrate InterCode's viability as a testbed by evaluating multiple state-of-the-art LLMs configured with different prompting strategies such as ReAct and Plan & Solve. Our results showcase the benefits of interactive code generation and demonstrate that InterCode can serve as a challenging benchmark for advancing code understanding and generation capabilities. InterCode is designed to be easily extensible and can even be used to create new tasks such as Capture the Flag, a popular coding puzzle that is inherently multi-step and involves multiple programming languages. Project site with code and data: https://intercode-benchmark.github.io
Large language models (LLMs) have been touted to enable increased productivity in many areas of today's work life. Scientific research as an area of work is no exception: the potential of LLM-based tools to assist in the daily work of scientists has become a highly discussed topic across disciplines. However, we are only at the very onset of this subject of study. It is still unclear how the potential of LLMs will materialise in research practice. With this study, we give first empirical evidence on the use of LLMs in the research process. We have investigated a set of use cases for LLM-based tools in scientific research, and conducted a first study to assess to which degree current tools are helpful. In this paper we report specifically on use cases related to software engineering, such as generating application code and developing scripts for data analytics. While we studied seemingly simple use cases, results across tools differ significantly. Our results highlight the promise of LLM-based tools in general, yet we also observe various issues, particularly regarding the integrity of the output these tools provide.
In this work, we develop and release Yuan 2.0, a series of large language models with parameters ranging from 2.1 billion to 102.6 billion. The Localized Filtering-based Attention (LFA) is introduced to incorporate prior knowledge of local dependencies of natural language into Attention. A data filtering and generating system is presented to build pre-training and fine-tuning dataset in high quality. A distributed training method with non-uniform pipeline parallel, data parallel, and optimizer parallel is proposed, which greatly reduces the bandwidth requirements of intra-node communication, and achieves good performance in large-scale distributed training. Yuan 2.0 models display impressive ability in code generation, math problem-solving, and chatting compared with existing models. The latest version of YUAN 2.0, including model weights and source code, is accessible at Github.
The recent explosion in the capabilities of large language models has led to a wave of interest in how best to prompt a model to perform a given task. While it may be tempting to simply choose a prompt based on average performance on a validation set, this can lead to a deployment where unexpectedly poor responses are generated, especially for the worst-off users. To mitigate this prospect, we propose Prompt Risk Control, a lightweight framework for selecting a prompt based on rigorous upper bounds on families of informative risk measures. We offer methods for producing bounds on a diverse set of metrics, including quantities that measure worst-case responses and disparities in generation quality across the population of users. In addition, we extend the underlying statistical bounding techniques to accommodate the possibility of distribution shifts in deployment. Experiments on applications such as open-ended chat, medical question summarization, and code generation highlight how such a framework can foster responsible deployment by reducing the risk of the worst outcomes.
Language models can serve as a valuable tool for software developers to increase productivity. Large generative models can be used for code generation and code completion, while smaller encoder-only models are capable of performing code search tasks using natural language queries.These capabilities are heavily influenced by the quality and diversity of the available training data. Source code datasets used for training usually focus on the most popular languages and testing is mostly conducted on the same distributions, often overlooking low-resource programming languages. Motivated by the NLP generalization taxonomy proposed by Hupkes et.\,al., we propose a new benchmark dataset called GenCodeSearchNet (GeCS) which builds upon existing natural language code search datasets to systemically evaluate the programming language understanding generalization capabilities of language models. As part of the full dataset, we introduce a new, manually curated subset StatCodeSearch that focuses on R, a popular but so far underrepresented programming language that is often used by researchers outside the field of computer science. For evaluation and comparison, we collect several baseline results using fine-tuned BERT-style models and GPT-style large language models in a zero-shot setting.
Contemporary Large Language Models (LLMs) exhibit a high degree of code generation and comprehension capability. A particularly promising area is their ability to interpret code modules from unfamiliar libraries for solving user-instructed tasks. Recent work has shown that large proprietary LLMs can learn novel library usage in-context from demonstrations. These results raise several open questions: whether demonstrations of library usage is required, whether smaller (and more open) models also possess such capabilities, etc. In this work, we take a broader approach by systematically evaluating a diverse array of LLMs across three scenarios reflecting varying levels of domain specialization to understand their abilities and limitations in generating code based on libraries defined in-context. Our results show that even smaller open-source LLMs like Llama-2 and StarCoder demonstrate an adept understanding of novel code libraries based on specification presented in-context. Our findings further reveal that LLMs exhibit a surprisingly high proficiency in learning novel library modules even when provided with just natural language descriptions or raw code implementations of the functions, which are often cheaper to obtain than demonstrations. Overall, our results pave the way for harnessing LLMs in more adaptable and dynamic coding environments.
This paper introduces INTERVENOR (INTERactiVE chaiN Of Repair), a system designed to emulate the interactive code repair processes observed in humans, encompassing both code diagnosis and code repair. INTERVENOR prompts Large Language Models (LLMs) to play distinct roles during the code repair process, functioning as both a Code Learner and a Code Teacher. Specifically, the Code Learner is tasked with adhering to instructions to generate or repair code, while the Code Teacher is responsible for crafting a Chain-of-Repair (CoR) to serve as guidance for the Code Learner. During generating the CoR, the Code Teacher needs to check the generated codes from Code Learner and reassess how to address code bugs based on error feedback received from compilers. Experimental results demonstrate that INTERVENOR surpasses baseline models, exhibiting improvements of approximately 18% and 4.3% over GPT-3.5 in code generation and code translation tasks, respectively. Our further analyses show that CoR is effective to illuminate the reasons behind bugs and outline solution plans in natural language. With the feedback of code compilers, INTERVENOR can accurately identify syntax errors and assertion errors and provide precise instructions to repair codes. All data and codes are available at https://github.com/NEUIR/INTERVENOR
Despite Large Language Models (LLMs) like GPT-4 achieving impressive results in function-level code generation, they struggle with repository-scale code understanding (e.g., coming up with the right arguments for calling routines), requiring a deeper comprehension of complex file interactions. Also, recently, people have developed LLM agents that attempt to interact with repository code (e.g., compiling and evaluating its execution), prompting the need to evaluate their performance. These gaps have motivated our development of ML-Bench, a benchmark rooted in real-world programming applications that leverage existing code repositories to perform tasks. Addressing the need for LLMs to interpret long code contexts and translate instructions into precise, executable scripts, ML-Bench encompasses annotated 9,641 examples across 18 GitHub repositories, challenging LLMs to accommodate user-specified arguments and documentation intricacies effectively. To evaluate both LLMs and AI agents, two setups are employed: ML-LLM-Bench for assessing LLMs' text-to-code conversion within a predefined deployment environment, and ML-Agent-Bench for testing autonomous agents in an end-to-end task execution within a Linux sandbox environment. Our findings indicate that while GPT-4o leads with a Pass@5 rate surpassing 50%, there remains significant scope for improvement, highlighted by issues such as hallucinated outputs and difficulties with bash script generation. Notably, in the more demanding ML-Agent-Bench, GPT-4o achieves a 76.47% success rate, reflecting the efficacy of iterative action and feedback in complex task resolution. Our code, dataset, and models are available at https://github.com/gersteinlab/ML-bench.
In this work, we present the largest benchmark to date on linguistic acceptability: Multilingual Evaluation of Linguistic Acceptability -- MELA, with 46K samples covering 10 languages from a diverse set of language families. We establish LLM baselines on this benchmark, and investigate cross-lingual transfer in acceptability judgements with XLM-R. In pursuit of multilingual interpretability, we conduct probing experiments with fine-tuned XLM-R to explore the process of syntax capability acquisition. Our results show that GPT-4o exhibits a strong multilingual ability, outperforming fine-tuned XLM-R, while open-source multilingual models lag behind by a noticeable gap. Cross-lingual transfer experiments show that transfer in acceptability judgment is non-trivial: 500 Icelandic fine-tuning examples lead to 23 MCC performance in a completely unrelated language -- Chinese. Results of our probing experiments indicate that training on MELA improves the performance of XLM-R on syntax-related tasks. Our data is available at https://github.com/sjtu-compling/MELA.
Large Language Models (LLMs) have demonstrated remarkable performance on assisting humans in programming and facilitating programming automation. However, existing benchmarks for evaluating the code understanding and generation capacities of LLMs suffer from severe limitations. First, most benchmarks are insufficient as they focus on a narrow range of popular programming languages and specific tasks, whereas real-world software development scenarios show a critical need to implement systems with multilingual and multitask programming environments to satisfy diverse requirements. Second, most benchmarks fail to consider the actual executability and the consistency of execution results of the generated code. To bridge these gaps between existing benchmarks and expectations from practical applications, we introduce CodeScope, an execution-based, multilingual, multitask, multidimensional evaluation benchmark for comprehensively measuring LLM capabilities on coding tasks. CodeScope covers 43 programming languages and eight coding tasks. It evaluates the coding performance of LLMs from three dimensions (perspectives): length, difficulty, and efficiency. To facilitate execution-based evaluations of code generation, we develop MultiCodeEngine, an automated code execution engine that supports 14 programming languages. Finally, we systematically evaluate and analyze eight mainstream LLMs and demonstrate the superior breadth and challenges of CodeScope for evaluating LLMs on code understanding and generation tasks compared to other benchmarks. The CodeScope benchmark and code are publicly available at https://github.com/WeixiangYAN/CodeScope.
Large language models (LLMs) such as GPT-3.5 and CodeLlama are powerful models for code generation and understanding. Fine-tuning these models comes with a high computational cost and requires a large labeled dataset. Alternatively, in-context learning techniques allow models to learn downstream tasks with only a few examples. Recently, researchers have shown how in-context learning performs well in bug detection and repair. In this paper, we propose code-pair classification task in which both the buggy and non-buggy versions are given to the model, and the model identifies the buggy ones. We evaluate our task in real-world dataset of bug detection and two most powerful LLMs. Our experiments indicate that an LLM can often pick the buggy from the non-buggy version of the code, and the code-pair classification task is much easier compared to be given a snippet and deciding if and where a bug exists.
Large Language Models (LLMs) recently demonstrated capabilities for generating source code in common programming languages. Additionally, commercial products such as ChatGPT 4 started to provide code interpreters, allowing for the automatic execution of generated code fragments, instant feedback, and the possibility to develop and refine in a conversational fashion. With an exploratory research approach, this paper applies code generation and interpretation to conceptual models. The concept and prototype of a conceptual model interpreter is explored, capable of rendering visual models generated in textual syntax by state-of-the-art LLMs such as Llama~2 and ChatGPT 4. In particular, these LLMs can generate textual syntax for the PlantUML and Graphviz modeling software that is automatically rendered within a conversational user interface. The first result is an architecture describing the components necessary to interact with interpreters and LLMs through APIs or locally, providing support for many commercial and open source LLMs and interpreters. Secondly, experimental results for models generated with ChatGPT 4 and Llama 2 are discussed in two cases covering UML and, on an instance level, graphs created from custom data. The results indicate the possibility of modeling iteratively in a conversational fashion.
Modular neural networks without additional training have recently been shown to surpass end-to-end neural networks on challenging vision-language tasks. The latest such methods simultaneously introduce LLM-based code generation to build programs and a number of skill-specific, task-oriented modules to execute them. In this paper, we focus on ViperGPT and ask where its additional performance comes from and how much is due to the (state-of-art, end-to-end) BLIP-2 model it subsumes vs. additional symbolic components. To do so, we conduct a controlled study (comparing end-to-end, modular, and prompting-based methods across several VQA benchmarks). We find that ViperGPT's reported gains over BLIP-2 can be attributed to its selection of task-specific modules, and when we run ViperGPT using a more task-agnostic selection of modules, these gains go away. Additionally, ViperGPT retains much of its performance if we make prominent alterations to its selection of modules: e.g. removing or retaining only BLIP-2. Finally, we compare ViperGPT against a prompting-based decomposition strategy and find that, on some benchmarks, modular approaches significantly benefit by representing subtasks with natural language, instead of code.
Code Large Language Models (CodeLLMs) have ushered in a new era in code generation advancements. However, selecting the best code solutions from all possible CodeLLM outputs remains a challenge. Previous methods often overlooked the intricate functional similarities and interactions between solution clusters. We introduce SRank, a novel reranking strategy for selecting the best solutions from code generation, focusing on modeling the relationships between clusters of solutions. By quantifying the functional overlap between solution clusters, our approach provides a better ranking strategy for code solutions. Empirical results show that our method achieves remarkable results on the pass@1 score. For instance, on the Human-Eval benchmark, we achieve 69.66% in pass@1 with Codex002, 75.31% with WizardCoder, 53.99% with StarCoder, and 60.55% with CodeGen, surpassing state-of-the-art code generation reranking methods such as CodeT and Coder-Reviewer on the same CodeLLM by a significant margin (approximately 6.1% improvement on average). Even in scenarios with a limited number of sampled solutions and test cases, our approach demonstrates robustness and superiority, marking a new benchmark in code generation reranking. Our implementation can be found at https://github.com/FSoft-AI4Code/SRank-CodeRanker.
This paper presents a comprehensive evaluation of the code generation capabilities of ChatGPT, a prominent large language model, compared to human programmers. A novel dataset of 131 code-generation prompts across 5 categories was curated to enable robust analysis. Code solutions were generated by both ChatGPT and humans for all prompts, resulting in 262 code samples. A meticulous manual assessment methodology prioritized evaluating correctness, comprehensibility, and security using 14 established code quality metrics. The key findings reveal ChatGPT's strengths in crafting concise, efficient code with advanced constructs, showcasing strengths in data analysis tasks (93.1% accuracy) but limitations in visual-graphical challenges. Comparative analysis with human code highlights ChatGPT's inclination towards modular design and superior error handling. Additionally, machine learning models effectively distinguished ChatGPT from human code with up to 88% accuracy, suggesting detectable coding style disparities. By providing profound insights into ChatGPT's code generation capabilities and limitations through quantitative metrics and qualitative analysis, this study makes valuable contributions toward advancing AI-based programming assistants. The curated dataset and methodology offer a robust foundation for future research in this nascent domain. All data and codes are available on https://github.com/DSAatUSU/ChatGPT-promises-and-pitfalls.
One major challenge of translating code between programming languages is that parallel training data is often limited. To overcome this challenge, we present two data augmentation techniques, one that builds comparable corpora (i.e., code pairs with similar functionality), and another that augments existing parallel data with multiple reference translations. Specifically, we build and analyze multiple types of comparable corpora, including programs generated from natural language documentation using a code generation model. Furthermore, to reduce overfitting to a single reference translation, we automatically generate additional translation references for available parallel data and filter the translations by unit tests, which increases variation in target translations. Experiments show that our data augmentation techniques significantly improve CodeT5 for translation between Java, Python, and C++ by an average of 7.5% Computational Accuracy (CA@1), which verifies the correctness of translations by execution. The code is available at https://github.com/Veronicium/CMTrans.
Instruction tuning is a burgeoning method to elicit the general intelligence of Large Language Models (LLMs). However, the creation of instruction data is still largely heuristic, leading to significant variation in quantity and quality across existing datasets. While some research advocates for expanding the number of instructions, others suggest that a small set of well-chosen examples is adequate. To better understand data construction guidelines, our research provides a granular analysis of how data volume, parameter size, and data construction methods influence the development of each underlying ability of LLM, such as creative writing, code generation, and logical reasoning. We present a meticulously curated dataset with over 40k instances across ten abilities and examine instruction-tuned models with 7b to 33b parameters. Our study reveals three primary findings: (i) Despite the models' overall performance being tied to data and parameter scale, individual abilities have different sensitivities to these factors. (ii) Human-curated data strongly outperforms synthetic data from GPT-4 in efficiency and can constantly enhance model performance with volume increases, but is unachievable with synthetic data. (iii) Instruction data brings powerful cross-ability generalization, as evidenced by out-of-domain evaluations. Furthermore, we demonstrate how these findings can guide more efficient data constructions, leading to practical performance improvements on two public benchmarks.
While large language models (LLMs) now excel at code generation, a key aspect of software development is the art of refactoring: consolidating code into libraries of reusable and readable programs. In this paper, we introduce LILO, a neurosymbolic framework that iteratively synthesizes, compresses, and documents code to build libraries tailored to particular problem domains. LILO combines LLM-guided program synthesis with recent algorithmic advances in automated refactoring from Stitch: a symbolic compression system that efficiently identifies optimal lambda abstractions across large code corpora. To make these abstractions interpretable, we introduce an auto-documentation (AutoDoc) procedure that infers natural language names and docstrings based on contextual examples of usage. In addition to improving human readability, we find that AutoDoc boosts performance by helping LILO's synthesizer to interpret and deploy learned abstractions. We evaluate LILO on three inductive program synthesis benchmarks for string editing, scene reasoning, and graphics composition. Compared to existing neural and symbolic methods - including the state-of-the-art library learning algorithm DreamCoder - LILO solves more complex tasks and learns richer libraries that are grounded in linguistic knowledge.
With the rise of powerful closed-sourced LLMs (ChatGPT, GPT-4), there are increasing interests in distilling the capabilies of close-sourced LLMs to smaller open-sourced LLMs. Previous distillation methods usually prompt ChatGPT to generate a set of instructions and answers, for the student model to learn. However, such standard distillation approach neglects the merits and conditions of the student model. Inspired by modern teaching principles, we design a personalised distillation process, in which the student attempts to solve a task first, then the teacher provides an adaptive refinement for the student to improve. Instead of feeding the student with teacher's prior, personalised distillation enables personalised learning for the student model, as it only learns on examples it makes mistakes upon and learns to improve its own solution. On code generation, personalised distillation consistently outperforms standard distillation with only one third of the data. With only 2.5-3K personalised examples that incur a data-collection cost of 4-6$, we boost CodeGen-mono-16B by 7% to achieve 36.4% pass@1 and StarCoder by 12.2% to achieve 45.8% pass@1 on HumanEval.