Despite their impressive performance, large language models (LMs) still struggle with reliably generating complex output structures when not finetuned to follow the required output format exactly. To address this issue, grammar-constrained decoding (GCD) can be used to control the generation of LMs, guaranteeing that the output follows a given structure. Most existing GCD methods are, however, limited to specific tasks, such as parsing or code generation. In this work, we demonstrate that formal grammars can describe the output space for a much wider range of tasks and argue that GCD can serve as a unified framework for structured NLP tasks in general. For increased flexibility, we introduce input-dependent grammars, which allow the grammar to depend on the input and thus enable the generation of different output structures for different inputs. We then empirically demonstrate the power and flexibility of GCD-enhanced LMs on (1) information extraction, (2) entity disambiguation, and (3) constituency parsing. Our results indicate that grammar-constrained LMs substantially outperform unconstrained LMs or even beat task-specific finetuned models. Grammar constraints thus hold great promise for harnessing off-the-shelf LMs for a wide range of structured NLP tasks, especially where training data is scarce or finetuning is expensive. Code and data: https://github.com/epfl-dlab/GCD.
While large language models (LLMs) bring not only performance but also complexity, recent work has started to turn LLMs into data generators rather than task inferencers, where another affordable task model is trained for efficient deployment and inference. However, such an approach has primarily been applied to natural language tasks and has not yet been explored for symbolic language tasks with complex structured outputs (e.g., semantic parsing and code generation). In this paper, we propose SymGen which utilizes LLMs for generating various annotation-expensive symbolic language data. SymGen consists of an informative prompt to steer generation and an agreement-based verifier to improve data correctness. We conduct extensive experiments on six symbolic language tasks across various settings. Compared with the LLMs, we demonstrate the 1\%-sized task model can achieve comparable or better performance, largely cutting inference and deployment costs. We also show that generated data with only a few human demonstrations can be as effective as over 10 times the amount of human-annotated data when training the task model, saving a considerable amount of annotation effort. SymGen sheds new light on data generation for complex tasks, and we release the code at \href{https://github.com/HKUNLP/SymGen}{https://github.com/HKUNLP/SymGen}.
A popular approach for improving the correctness of output from large language models (LLMs) is Self-Consistency - poll the LLM multiple times and output the most frequent solution. Existing Self-Consistency techniques always generate a constant number of samples per question, where a better approach will be to non-uniformly distribute the available budget based on the amount of agreement in the samples generated so far. In response, we introduce Adaptive-Consistency, a cost-efficient, model-agnostic technique that dynamically adjusts the number of samples per question using a lightweight stopping criterion. Our experiments over 17 reasoning and code generation datasets and three LLMs demonstrate that Adaptive-Consistency reduces sample budget by up to 7.9 times with an average accuracy drop of less than 0.1%. Our code and data are available at https://www.sample-step-by-step.info
Automatic code generation has recently attracted large attention and is becoming more significant to the software development process. Solutions based on Machine Learning and Artificial Intelligence are being used to increase human and software efficiency in potent and innovative ways. In this paper, we aim to leverage these developments and introduce a novel approach to generating frontend component code for the popular Angular framework. We propose to do this using behavior-driven development test specifications as input to a transformer-based machine learning model. Our approach aims to drastically reduce the development time needed for web applications while potentially increasing software quality and introducing new research ideas toward automatic code generation.
Dynamic control flow is an important technique often used to design expressive and efficient deep learning computations for applications such as text parsing, machine translation, exiting early out of deep models and so on. The control flow divergence resulting from dynamic control flow makes batching, an important optimization enabling high throughput and hardware utilization, difficult to perform manually. In this paper, we present ACRoBat, a framework that enables efficient automatic batching for dynamic deep learning computations by performing hybrid static+dynamic compiler optimizations and end-to-end tensor code generation. ACRoBat performs up to 8.5X better than DyNet, a state-of-the-art framework for automatic batching, on an Nvidia GeForce GPU.
Code generation aims to automatically generate source code from high-level task specifications, which can significantly increase productivity of software engineering. Recently, approaches based on large language models (LLMs) have shown remarkable code generation abilities on simple tasks. However, generate code for more complex tasks, such as competition-level problems, remains challenging. In this paper, we introduce Brainstorm framework for code generation. It leverages a brainstorming step that generates and selects diverse thoughts on the problem to facilitate algorithmic reasoning, where the thoughts are possible blueprint of solving the problem. We demonstrate that Brainstorm significantly enhances the ability of LLMs to solve competition-level programming problems, resulting in a more than 50% increase in the pass@$k$ metrics for ChatGPT on the CodeContests benchmark, achieving state-of-the-art performance. Furthermore, our experiments conducted on LeetCode contests show that our framework boosts the ability of ChatGPT to a level comparable to that of human programmers.
We introduce PaLM 2, a new state-of-the-art language model that has better multilingual and reasoning capabilities and is more compute-efficient than its predecessor PaLM. PaLM 2 is a Transformer-based model trained using a mixture of objectives. Through extensive evaluations on English and multilingual language, and reasoning tasks, we demonstrate that PaLM 2 has significantly improved quality on downstream tasks across different model sizes, while simultaneously exhibiting faster and more efficient inference compared to PaLM. This improved efficiency enables broader deployment while also allowing the model to respond faster, for a more natural pace of interaction. PaLM 2 demonstrates robust reasoning capabilities exemplified by large improvements over PaLM on BIG-Bench and other reasoning tasks. PaLM 2 exhibits stable performance on a suite of responsible AI evaluations, and enables inference-time control over toxicity without additional overhead or impact on other capabilities. Overall, PaLM 2 achieves state-of-the-art performance across a diverse set of tasks and capabilities. When discussing the PaLM 2 family, it is important to distinguish between pre-trained models (of various sizes), fine-tuned variants of these models, and the user-facing products that use these models. In particular, user-facing products typically include additional pre- and post-processing steps. Additionally, the underlying models may evolve over time. Therefore, one should not expect the performance of user-facing products to exactly match the results reported in this report.
Fine-tuning pre-trained language models (LMs) is essential for enhancing their capabilities. Existing techniques commonly fine-tune on input-output pairs (e.g., instruction tuning) or with numerical rewards that gauge the output quality (e.g., RLHF). We explore LMs' potential to learn from textual interactions (LETI) that not only check their correctness with binary labels but also pinpoint and explain errors in their outputs through textual feedback. Our focus is the code generation task, where the model produces code based on natural language instructions. This setting invites a natural and scalable way to acquire textual feedback: the error messages and stack traces from code execution using a Python interpreter. LETI iteratively fine-tunes the model, using the LM objective, on a concatenation of natural language instructions, LM-generated programs, and textual feedback. Prepended to this fine-tuning text, a binary reward token is used to differentiate correct and buggy solutions. LETI requires no ground-truth outputs for training and even outperforms a fine-tuned baseline that does. LETI not only improves the performance of LMs on a code generation dataset MBPP, but also generalizes to other datasets. Trained on MBPP, it achieves comparable or better performance than the base LMs on unseen problems in HumanEval. Furthermore, compared to binary feedback, we observe that textual feedback leads to improved generation quality and sample efficiency, achieving the same performance with fewer than half of the gradient steps. LETI is equally applicable in natural language tasks when they can be formulated as code generation, which we empirically verified on event argument extraction.
This paper evaluates the capability of two state-of-the-art artificial intelligence (AI) models, GPT-3.5 and Bard, in generating Java code given a function description. We sourced the descriptions from CodingBat.com, a popular online platform that provides practice problems to learn programming. We compared the Java code generated by both models based on correctness, verified through the platform's own test cases. The results indicate clear differences in the capabilities of the two models. GPT-3.5 demonstrated superior performance, generating correct code for approximately 90.6% of the function descriptions, whereas Bard produced correct code for 53.1% of the functions. While both models exhibited strengths and weaknesses, these findings suggest potential avenues for the development and refinement of more advanced AI-assisted code generation tools. The study underlines the potential of AI in automating and supporting aspects of software development, although further research is required to fully realize this potential.
Message Passing Interface (MPI) plays a crucial role in distributed memory parallelization across multiple nodes. However, parallelizing MPI code manually, and specifically, performing domain decomposition, is a challenging, error-prone task. In this paper, we address this problem by developing MPI-RICAL, a novel data-driven, programming-assistance tool that assists programmers in writing domain decomposition based distributed memory parallelization code. Specifically, we train a supervised language model to suggest MPI functions and their proper locations in the code on the fly. We also introduce MPICodeCorpus, the first publicly available corpus of MPI-based parallel programs that is created by mining more than 15,000 open-source repositories on GitHub. Experimental results have been done on MPICodeCorpus and more importantly, on a compiled benchmark of MPI-based parallel programs for numerical computations that represent real-world scientific applications. MPI-RICAL achieves F1 scores between 0.87-0.91 on these programs, demonstrating its accuracy in suggesting correct MPI functions at appropriate code locations.. The source code used in this work, as well as other relevant sources, are available at: https://github.com/Scientific-Computing-Lab-NRCN/MPI-rical
Large language models (LLMs) pretrained on vast source code have achieved prominent progress in code intelligence. However, existing code LLMs have two main limitations in terms of architecture and pretraining tasks. First, they often adopt a specific architecture (encoder-only or decoder-only) or rely on a unified encoder-decoder network for different downstream tasks. The former paradigm is limited by inflexibility in applications while in the latter, the model is treated as a single system for all tasks, leading to suboptimal performance on a subset of tasks. Secondly, they often employ a limited set of pretraining objectives which might not be relevant to some downstream tasks and hence result in substantial performance degrade. To address these limitations, we propose ``CodeT5+'', a family of encoder-decoder LLMs for code in which component modules can be flexibly combined to suit a wide range of downstream code tasks. Such flexibility is enabled by our proposed mixture of pretraining objectives to mitigate the pretrain-finetune discrepancy. These objectives cover span denoising, contrastive learning, text-code matching, and causal LM pretraining tasks, on both unimodal and bimodal multilingual code corpora. Furthermore, we propose to initialize CodeT5+ with frozen off-the-shelf LLMs without training from scratch to efficiently scale up our models, and explore instruction-tuning to align with natural language instructions. We extensively evaluate CodeT5+ on over 20 code-related benchmarks in different settings, including zero-shot, finetuning, and instruction-tuning. We observe state-of-the-art (SoTA) model performance on various code-related tasks, such as code generation and completion, math programming, and text-to-code retrieval tasks. Particularly, our instruction-tuned CodeT5+ 16B achieves new SoTA results on HumanEval code generation task against other open code LLMs.
The increasing popularity of large language models (LLMs) such as ChatGPT has led to growing concerns about their safety, security risks, and ethical implications. This paper aims to provide an overview of the different types of security risks associated with ChatGPT, including malicious text and code generation, private data disclosure, fraudulent services, information gathering, and producing unethical content. We present an empirical study examining the effectiveness of ChatGPT's content filters and explore potential ways to bypass these safeguards, demonstrating the ethical implications and security risks that persist in LLMs even when protections are in place. Based on a qualitative analysis of the security implications, we discuss potential strategies to mitigate these risks and inform researchers, policymakers, and industry professionals about the complex security challenges posed by LLMs like ChatGPT. This study contributes to the ongoing discussion on the ethical and security implications of LLMs, underscoring the need for continued research in this area.
Large Language Models (LLMs) (e.g., ChatGPT) have shown impressive performance in code generation. LLMs take prompts as inputs, and Chain-of-Thought (CoT) prompting is the state-of-the-art prompting technique. CoT prompting asks LLMs first to generate CoTs (i.e., intermediate natural language reasoning steps) and then output the code. However, CoT prompting is designed for natural language generation and has low accuracy in code generation. In this paper, we propose Structured CoTs (SCoTs) and present a novel prompting technique for code generation, named SCoT prompting. Our motivation is source code contains rich structural information and any code can be composed of three program structures (i.e., sequence, branch, and loop structures). Intuitively, structured intermediate reasoning steps make for structured source code. Thus, we ask LLMs to use program structures to build CoTs, obtaining SCoTs. Then, LLMs generate the final code based on SCoTs. Compared to CoT prompting, SCoT prompting explicitly constrains LLMs to think about how to solve requirements from the view of source code and further the performance of LLMs in code generation. We apply SCoT prompting to two LLMs (i.e., ChatGPT and Codex) and evaluate it on three benchmarks (i.e., HumanEval, MBPP, and MBCPP). (1) SCoT prompting outperforms the state-of-the-art baseline - CoT prompting by up to 13.79% in Pass@1. (2) Human evaluation shows human developers prefer programs from SCoT prompting. (3) SCoT prompting is robust to examples and achieves substantial improvements.
Large Language Models (LLMs), such as ChatGPT, demonstrate a strong understanding of human natural language and have been explored and applied in various fields, including reasoning, creative writing, code generation, translation, and information retrieval. By adopting LLM as the reasoning core, we introduce Autonomous GIS as an AI-powered geographic information system (GIS) that leverages the LLM's general abilities in natural language understanding, reasoning, and coding for addressing spatial problems with automatic spatial data collection, analysis, and visualization. We envision that autonomous GIS will need to achieve five autonomous goals: self-generating, self-organizing, self-verifying, self-executing, and self-growing. We developed a prototype system called LLM-Geo using the GPT-4 API in a Python environment, demonstrating what an autonomous GIS looks like and how it delivers expected results without human intervention using three case studies. For all case studies, LLM-Geo was able to return accurate results, including aggregated numbers, graphs, and maps, significantly reducing manual operation time. Although still in its infancy and lacking several important modules such as logging and code testing, LLM-Geo demonstrates a potential path toward the next-generation AI-powered GIS. We advocate for the GIScience community to dedicate more effort to the research and development of autonomous GIS, making spatial analysis easier, faster, and more accessible to a broader audience.
The BigCode community, an open-scientific collaboration working on the responsible development of Large Language Models for Code (Code LLMs), introduces StarCoder and StarCoderBase: 15.5B parameter models with 8K context length, infilling capabilities and fast large-batch inference enabled by multi-query attention. StarCoderBase is trained on 1 trillion tokens sourced from The Stack, a large collection of permissively licensed GitHub repositories with inspection tools and an opt-out process. We fine-tuned StarCoderBase on 35B Python tokens, resulting in the creation of StarCoder. We perform the most comprehensive evaluation of Code LLMs to date and show that StarCoderBase outperforms every open Code LLM that supports multiple programming languages and matches or outperforms the OpenAI code-cushman-001 model. Furthermore, StarCoder outperforms every model that is fine-tuned on Python, can be prompted to achieve 40\% pass@1 on HumanEval, and still retains its performance on other programming languages. We take several important steps towards a safe open-access model release, including an improved PII redaction pipeline and a novel attribution tracing tool, and make the StarCoder models publicly available under a more commercially viable version of the Open Responsible AI Model license.
We present The Vault, a dataset of high-quality code-text pairs in multiple programming languages for training large language models to understand and generate code. We present methods for thoroughly extracting samples that use both rule-based and deep learning-based methods to ensure that they contain high-quality pairs of code and text, resulting in a dataset of 43 million high-quality code-text pairs. Our extensive evaluations on common coding tasks including code generation, code search and code summarization show that when fine-tuning Code Large Language Models on The Vault, such models outperform the same models trained on other datasets such as CodeSearchNet. We also provide detailed analyses of our datasets to assess the effects of various programming languages and docstrings on the performance of such models.
Large language models (LLMs) pre-trained on massive corpora have demonstrated impressive few-shot learning ability on many NLP tasks. A common practice is to recast the task into a text-to-text format such that generative LLMs of natural language (NL-LLMs) like GPT-3 can be prompted to solve it. However, it is nontrivial to perform information extraction (IE) tasks with NL-LLMs since the output of the IE task is usually structured and therefore is hard to be converted into plain text. In this paper, we propose to recast the structured output in the form of code instead of natural language and utilize generative LLMs of code (Code-LLMs) such as Codex to perform IE tasks, in particular, named entity recognition and relation extraction. In contrast to NL-LLMs, we show that Code-LLMs can be well-aligned with these IE tasks by designing code-style prompts and formulating these IE tasks as code generation tasks. Experiment results on seven benchmarks show that our method consistently outperforms fine-tuning moderate-size pre-trained models specially designed for IE tasks (e.g., UIE) and prompting NL-LLMs under few-shot settings. We further conduct a series of in-depth analyses to demonstrate the merits of leveraging Code-LLMs for IE tasks.
Automated source code generation is currently a popular machine-learning-based task. It can be helpful for software developers to write functionally correct code from a given context. However, just like human developers, a code generation model can produce vulnerable code, which the developers can mistakenly use. For this reason, evaluating the security of a code generation model is a must. In this paper, we describe SecurityEval, an evaluation dataset to fulfill this purpose. It contains 130 samples for 75 vulnerability types, which are mapped to the Common Weakness Enumeration (CWE). We also demonstrate using our dataset to evaluate one open-source (i.e., InCoder) and one closed-source code generation model (i.e., GitHub Copilot).
Code execution is a fundamental aspect of programming language semantics that reflects the exact behavior of the code. However, most pre-trained models for code intelligence ignore the execution trace and only rely on source code and syntactic structures. In this paper, we investigate how well pre-trained models can understand and perform code execution. We develop a mutation-based data augmentation technique to create a large-scale and realistic Python dataset and task for code execution, which challenges existing models such as Codex. We then present CodeExecutor, a Transformer model that leverages code execution pre-training and curriculum learning to enhance its semantic comprehension. We evaluate CodeExecutor on code execution and show its promising performance and limitations. We also demonstrate its potential benefits for code intelligence tasks such as zero-shot code-to-code search and text-to-code generation. Our analysis provides insights into the learning and generalization abilities of pre-trained models for code execution.
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.