Modern language models (LMs) have been successfully employed in source code generation and understanding, leading to a significant increase in research focused on learning-based code intelligence, such as automated bug repair, and test case generation. Despite their great potential, language models for code intelligence (LM4Code) are susceptible to potential pitfalls, which hinder realistic performance and further impact their reliability and applicability in real-world deployment. Such challenges drive the need for a comprehensive understanding - not just identifying these issues but delving into their possible implications and existing solutions to build more reliable language models tailored to code intelligence. Based on a well-defined systematic research approach, we conducted an extensive literature review to uncover the pitfalls inherent in LM4Code. Finally, 67 primary studies from top-tier venues have been identified. After carefully examining these studies, we designed a taxonomy of pitfalls in LM4Code research and conducted a systematic study to summarize the issues, implications, current solutions, and challenges of different pitfalls for LM4Code systems. We developed a comprehensive classification scheme that dissects pitfalls across four crucial aspects: data collection and labeling, system design and learning, performance evaluation, and deployment and maintenance. Through this study, we aim to provide a roadmap for researchers and practitioners, facilitating their understanding and utilization of LM4Code in reliable and trustworthy ways.
The use of large language models for code generation is a rapidly growing trend in software development. However, without effective methods for ensuring the correctness of generated code, this trend could lead to undesirable outcomes. In this paper, we introduce a new approach for addressing this challenge: the Clover paradigm, short for Closed-Loop Verifiable Code Generation, which uses consistency checking to provide a strong filter for incorrect code. Clover performs consistency checks among code, docstrings, and formal annotations. The checker is implemented using a novel integration of formal verification tools and large language models. We provide a theoretical analysis to support our thesis that Clover should be effective at consistency checking. We also empirically investigate its performance on a hand-designed dataset (CloverBench) featuring annotated Dafny programs at a textbook level of difficulty. Experimental results show that for this dataset: (i) LLMs are reasonably successful at automatically generating formal specifications; and (ii) our consistency checker achieves a promising acceptance rate (up to 87%) for correct instances while maintaining zero tolerance for adversarial incorrect ones (no false positives). Clover also discovered 6 incorrect programs in the existing human-written dataset MBPP-DFY-50.
Large language models (LLMs) excel at processing and generating both text and code. However, LLMs have had limited applicability in grounded task-oriented dialogue as they are difficult to steer toward task objectives and fail to handle novel grounding. We present a modular and interpretable grounded dialogue system that addresses these shortcomings by composing LLMs with a symbolic planner and grounded code execution. Our system consists of a reader and planner: the reader leverages an LLM to convert partner utterances into executable code, calling functions that perform grounding. The translated code's output is stored to track dialogue state, while a symbolic planner determines the next appropriate response. We evaluate our system's performance on the demanding OneCommon dialogue task, involving collaborative reference resolution on abstract images of scattered dots. Our system substantially outperforms the previous state-of-the-art, including improving task success in human evaluations from 56% to 69% in the most challenging setting.
Spiking neural networks (SNNs) aim to realize brain-inspired intelligence on neuromorphic chips with high energy efficiency by introducing neural dynamics and spike properties. As the emerging spiking deep learning paradigm attracts increasing interest, traditional programming frameworks cannot meet the demands of the automatic differentiation, parallel computation acceleration, and high integration of processing neuromorphic datasets and deployment. In this work, we present the SpikingJelly framework to address the aforementioned dilemma. We contribute a full-stack toolkit for pre-processing neuromorphic datasets, building deep SNNs, optimizing their parameters, and deploying SNNs on neuromorphic chips. Compared to existing methods, the training of deep SNNs can be accelerated $11\times$, and the superior extensibility and flexibility of SpikingJelly enable users to accelerate custom models at low costs through multilevel inheritance and semiautomatic code generation. SpikingJelly paves the way for synthesizing truly energy-efficient SNN-based machine intelligence systems, which will enrich the ecology of neuromorphic computing.
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.
With the recent focus on Large Language Models (LLMs), both StarCoder (Li et al., 2023) and Code Llama (Rozi\`ere et al., 2023) have demonstrated remarkable performance in code generation. However, there is still a need for improvement in code translation functionality with efficient training techniques. In response to this, we introduce SteloCoder, a decoder-only StarCoder-based LLM designed specifically for multi-programming language-to-Python code translation. In particular, SteloCoder achieves C++, C#, JavaScript, Java, or PHP-to-Python code translation without specifying the input programming language. We modified StarCoder model architecture by incorporating a Mixture-of-Experts (MoE) technique featuring five experts and a gating network for multi-task handling. Experts are obtained by StarCoder fine-tuning. Specifically, we use a Low-Rank Adaptive Method (LoRA) technique, limiting each expert size as only 0.06% of number of StarCoder's parameters. At the same time, to enhance training efficiency in terms of time, we adopt curriculum learning strategy and use self-instruct data for efficient fine-tuning. As a result, each expert takes only 6 hours to train on one single 80Gb A100 HBM. With experiments on XLCoST datasets, SteloCoder achieves an average of 73.76 CodeBLEU score in multi-programming language-to-Python translation, surpassing the top performance from the leaderboard by at least 3.5. This accomplishment is attributed to only 45M extra parameters with StarCoder as the backbone and 32 hours of valid training on one 80GB A100 HBM. The source code is release here: https://github.com/sade-adrien/SteloCoder.
Code Large Language Models (Code LLMs) are being increasingly employed in real-life applications, so evaluating them is critical. While the conventional accuracy evaluates the performance of Code LLMs on a set of individual tasks, their self-consistency across different tasks is overlooked. Intuitively, a trustworthy model should be self-consistent when generating natural language specifications for its own code and generating code for its own specifications. Failure to preserve self-consistency reveals a lack of understanding of the shared semantics underlying natural language and programming language, and therefore undermines the trustworthiness of a model. In this paper, we first formally define the self-consistency of Code LLMs and then design a framework, IdentityChain, which effectively and efficiently evaluates the self-consistency and conventional accuracy of a model at the same time. We study eleven Code LLMs and show that they fail to preserve self-consistency, which is indeed a distinct aspect from conventional accuracy. Furthermore, we show that IdentityChain can be used as a model debugging tool to expose weaknesses of Code LLMs by demonstrating three major weaknesses that we identify in current models using IdentityChain. Our code is available at https://github.com/marcusm117/IdentityChain.
The advent of large pre-trained language models in the domain of Code Synthesis has shown remarkable performance on various benchmarks, treating the problem of Code Generation in a fashion similar to Natural Language Generation, trained with a Language Modelling (LM) objective. In addition, the property of programming language code being precisely evaluable with respect to its semantics -- through the use of Unit Tests to check its functional correctness -- lends itself to using Reinforcement Learning (RL) as a further training paradigm. Previous work has shown that RL can be applied as such to improve models' coding capabilities; however, such RL-based methods rely on a reward signal based on defined Unit Tests, which are much harder to obtain compared to the huge crawled code datasets used in LM objectives. In this work, we present a novel approach to automatically obtain data consisting of function signatures and associated Unit Tests, suitable for RL training of Code Synthesis models. We also introduce a straightforward, simple yet effective Actor-Critic RL training scheme and show that it, in conjunction with automatically generated training data, leads to improvement of a pre-trained code language model's performance by up to 9.9% improvement over the original underlying code synthesis LM, and up to 4.3% over RL-based models trained with standard PPO or CodeRL.
Large Language Models (LLMs) have already become quite proficient at solving simpler programming tasks like those in HumanEval or MBPP benchmarks. However, solving more complex and competitive programming tasks is still quite challenging for these models - possibly due to their tendency to generate solutions as monolithic code blocks instead of decomposing them into logical sub-tasks and sub-modules. On the other hand, experienced programmers instinctively write modularized code with abstraction for solving complex tasks, often reusing previously developed modules. To address this gap, we propose CodeChain, a novel framework for inference that elicits modularized code generation through a chain of self-revisions, each being guided by some representative sub-modules generated in previous iterations. Concretely, CodeChain first instructs the LLM to generate modularized codes through chain-of-thought prompting. Then it applies a chain of self-revisions by iterating the two steps: 1) extracting and clustering the generated sub-modules and selecting the cluster representatives as the more generic and re-usable implementations, and 2) augmenting the original chain-of-thought prompt with these selected module-implementations and instructing the LLM to re-generate new modularized solutions. We find that by naturally encouraging the LLM to reuse the previously developed and verified sub-modules, CodeChain can significantly boost both modularity as well as correctness of the generated solutions, achieving relative pass@1 improvements of 35% on APPS and 76% on CodeContests. It is shown to be effective on both OpenAI LLMs as well as open-sourced LLMs like WizardCoder. We also conduct comprehensive ablation studies with different methods of prompting, number of clusters, model sizes, program qualities, etc., to provide useful insights that underpin CodeChain's success.
Large language models are a form of artificial intelligence systems whose primary knowledge consists of the statistical patterns, semantic relationships, and syntactical structures of language1. Despite their limited forms of "knowledge", these systems are adept at numerous complex tasks including creative writing, storytelling, translation, question-answering, summarization, and computer code generation. However, they have yet to demonstrate advanced applications in natural science. Here we show how large language models can perform scientific synthesis, inference, and explanation. We present a method for using general-purpose large language models to make inferences from scientific datasets of the form usually associated with special-purpose machine learning algorithms. We show that the large language model can augment this "knowledge" by synthesizing from the scientific literature. When a conventional machine learning system is augmented with this synthesized and inferred knowledge it can outperform the current state of the art across a range of benchmark tasks for predicting molecular properties. This approach has the further advantage that the large language model can explain the machine learning system's predictions. We anticipate that our framework will open new avenues for AI to accelerate the pace of scientific discovery.
Large vision-language models (VLMs) have achieved substantial progress in multimodal perception and reasoning. When integrated into an embodied agent, existing embodied VLM works either output detailed action sequences at the manipulation level or only provide plans at an abstract level, leaving a gap between high-level planning and real-world manipulation. To bridge this gap, we introduce Octopus, an embodied vision-language programmer that uses executable code generation as a medium to connect planning and manipulation. Octopus is designed to 1) proficiently comprehend an agent's visual and textual task objectives, 2) formulate intricate action sequences, and 3) generate executable code. To facilitate Octopus model development, we introduce OctoVerse: a suite of environments tailored for benchmarking vision-based code generators on a wide spectrum of tasks, ranging from mundane daily chores in simulators to sophisticated interactions in complex video games such as Grand Theft Auto (GTA) and Minecraft. To train Octopus, we leverage GPT-4 to control an explorative agent that generates training data, i.e., action blueprints and corresponding executable code. We also collect feedback that enables an enhanced training scheme called Reinforcement Learning with Environmental Feedback (RLEF). Through a series of experiments, we demonstrate Octopus's functionality and present compelling results, showing that the proposed RLEF refines the agent's decision-making. By open-sourcing our simulation environments, dataset, and model architecture, we aspire to ignite further innovation and foster collaborative applications within the broader embodied AI community.
The revolution of artificial intelligence content generation has been rapidly accelerated with the booming text-to-image (T2I) diffusion models. Within just two years of development, it was unprecedentedly of high-quality, diversity, and creativity that the state-of-the-art models could generate. However, a prevalent limitation persists in the effective communication with these popular T2I models, such as Stable Diffusion, using natural language descriptions. This typically makes an engaging image hard to obtain without expertise in prompt engineering with complex word compositions, magic tags, and annotations. Inspired by the recently released DALLE3 - a T2I model directly built-in ChatGPT that talks human language, we revisit the existing T2I systems endeavoring to align human intent and introduce a new task - interactive text to image (iT2I), where people can interact with LLM for interleaved high-quality image generation/edit/refinement and question answering with stronger images and text correspondences using natural language. In addressing the iT2I problem, we present a simple approach that augments LLMs for iT2I with prompting techniques and off-the-shelf T2I models. We evaluate our approach for iT2I in a variety of common-used scenarios under different LLMs, e.g., ChatGPT, LLAMA, Baichuan, and InternLM. We demonstrate that our approach could be a convenient and low-cost way to introduce the iT2I ability for any existing LLMs and any text-to-image models without any training while bringing little degradation on LLMs' inherent capabilities in, e.g., question answering and code generation. We hope this work could draw broader attention and provide inspiration for boosting user experience in human-machine interactions alongside the image quality of the next-generation T2I systems.
Aligned large language models (LLMs) demonstrate exceptional capabilities in task-solving, following instructions, and ensuring safety. However, the continual learning aspect of these aligned LLMs has been largely overlooked. Existing continual learning benchmarks lack sufficient challenge for leading aligned LLMs, owing to both their simplicity and the models' potential exposure during instruction tuning. In this paper, we introduce TRACE, a novel benchmark designed to evaluate continual learning in LLMs. TRACE consists of 8 distinct datasets spanning challenging tasks including domain-specific tasks, multilingual capabilities, code generation, and mathematical reasoning. All datasets are standardized into a unified format, allowing for effortless automatic evaluation of LLMs. Our experiments show that after training on TRACE, aligned LLMs exhibit significant declines in both general ability and instruction-following capabilities. For example, the accuracy of llama2-chat 13B on gsm8k dataset declined precipitously from 28.8\% to 2\% after training on our datasets. This highlights the challenge of finding a suitable tradeoff between achieving performance on specific tasks while preserving the original prowess of LLMs. Empirical findings suggest that tasks inherently equipped with reasoning paths contribute significantly to preserving certain capabilities of LLMs against potential declines. Motivated by this, we introduce the Reasoning-augmented Continual Learning (RCL) approach. RCL integrates task-specific cues with meta-rationales, effectively reducing catastrophic forgetting in LLMs while expediting convergence on novel tasks.
Counterfactual reasoning, a fundamental aspect of human cognition, involves contemplating alternatives to established facts or past events, significantly enhancing our abilities in planning and decision-making. In light of the advancements in current multi-modal large language models, we explore their effectiveness in counterfactual reasoning. To facilitate this investigation, we introduce a novel dataset, C-VQA, specifically designed to test the counterfactual reasoning capabilities of modern multi-modal large language models. This dataset is constructed by infusing original questions with counterfactual presuppositions, spanning various types such as numerical and boolean queries. It encompasses a mix of real and synthetic data, representing a wide range of difficulty levels. Our thorough evaluations of contemporary vision-language models using this dataset have revealed substantial performance drops, with some models showing up to a 40% decrease, highlighting a significant gap between current models and human-like vision reasoning capabilities. We hope our dataset will serve as a vital benchmark for evaluating the counterfactual reasoning capabilities of models. Code and dataset are publicly available at https://bzhao.me/C-VQA/.
We introduce Mistral 7B v0.1, a 7-billion-parameter language model engineered for superior performance and efficiency. Mistral 7B outperforms Llama 2 13B across all evaluated benchmarks, and Llama 1 34B in reasoning, mathematics, and code generation. Our model leverages grouped-query attention (GQA) for faster inference, coupled with sliding window attention (SWA) to effectively handle sequences of arbitrary length with a reduced inference cost. We also provide a model fine-tuned to follow instructions, Mistral 7B -- Instruct, that surpasses the Llama 2 13B -- Chat model both on human and automated benchmarks. Our models are released under the Apache 2.0 license.
Large language models (LLMs) are a new and powerful tool for a wide span of applications involving natural language and demonstrate impressive code generation abilities. The goal of this work is to automatically generate tests and use these tests to validate and verify compiler implementations of a directive-based parallel programming paradigm, OpenACC. To do so, in this paper, we explore the capabilities of state-of-the-art LLMs, including open-source LLMs -- Meta Codellama, Phind fine-tuned version of Codellama, Deepseek Deepseek Coder and closed-source LLMs -- OpenAI GPT-3.5-Turbo and GPT-4-Turbo. We further fine-tuned the open-source LLMs and GPT-3.5-Turbo using our own testsuite dataset along with using the OpenACC specification. We also explored these LLMs using various prompt engineering techniques that include code template, template with retrieval-augmented generation (RAG), one-shot example, one-shot with RAG, expressive prompt with code template and RAG. This paper highlights our findings from over 5000 tests generated via all the above mentioned methods. Our contributions include: (a) exploring the capabilities of the latest and relevant LLMs for code generation, (b) investigating fine-tuning and prompt methods, and (c) analyzing the outcome of LLMs generated tests including manually analysis of representative set of tests. We found the LLM Deepseek-Coder-33b-Instruct produced the most passing tests followed by GPT-4-Turbo.
Large language models (LLMs) with enormous pre-training tokens and parameters emerge diverse abilities, including math reasoning, code generation, and instruction following. These abilities are further enhanced by supervised fine-tuning (SFT). While the open-source community has explored ad-hoc SFT for enhancing individual capabilities, proprietary LLMs exhibit versatility across various skills. Therefore, understanding the facilitation of multiple abilities via SFT is paramount. In this study, we specifically focuses on the interplay of data composition between mathematical reasoning, code generation, and general human-aligning abilities during SFT. We propose four intriguing research questions to explore the association between model performance and various factors including data amount, composition ratio, model size and SFT strategies. Our experiments reveal that distinct capabilities scale differently and larger models generally show superior performance with same amount of data. Mathematical reasoning and code generation consistently improve with increasing data amount, whereas general abilities plateau after roughly a thousand samples. Moreover, we observe data composition appears to enhance various abilities under limited data conditions, yet can lead to performance conflicts when data is plentiful. Our findings also suggest the amount of composition data influences performance more than the composition ratio. In analysis of SFT strategies, we find that sequentially learning multiple skills risks catastrophic forgetting. Our proposed Dual-stage Mixed Fine-tuning (DMT) strategy offers a promising solution to learn multiple abilities with different scaling patterns.
While language models (LMs) have shown potential across a range of decision-making tasks, their reliance on simple acting processes limits their broad deployment as autonomous agents. In this paper, we introduce Language Agent Tree Search (LATS) -- the first general framework that synergizes the capabilities of LMs in reasoning, acting, and planning. By leveraging the in-context learning ability of LMs, we integrate Monte Carlo Tree Search into LATS to enable LMs as agents, along with LM-powered value functions and self-reflections for proficient exploration and enhanced decision-making. A key feature of our approach is the incorporation of an environment for external feedback, which offers a more deliberate and adaptive problem-solving mechanism that surpasses the constraints of existing techniques. Our experimental evaluation across diverse domains, including programming, interactive question-answering (QA), web navigation, and math, validates the effectiveness and generality of LATS in decision-making while maintaining competitive or improved reasoning performance. Notably, LATS achieves state-of-the-art pass@1 accuracy (92.7%) for programming on HumanEval with GPT-4 and demonstrates gradient-free performance (average score of 75.9) comparable to gradient-based fine-tuning for web navigation on WebShop with GPT-3.5. Code can be found at https://github.com/lapisrocks/LanguageAgentTreeSearch
Several recent advances in AI systems solve problems by providing a "scaffolding" program that structures multiple calls to language models (LMs) to generate better outputs. A scaffolding program is written in a programming language such as Python. In this work, we use a language-model-infused scaffolding program to improve itself. We start with a seed "improver" that improves an input program according to a given utility function by querying an LM several times and returning the best solution. We then run this seed improver to improve itself. Across a small set of downstream tasks, the resulting improved improver generates programs with significantly better performance than its seed improver. A variety of self-improvement strategies are proposed by the language model, including beam search, genetic algorithms, and simulated annealing. Since the language models themselves are not altered, this is not full recursive self-improvement. Nonetheless, it demonstrates that a modern language model, GPT-4 in our experiments, is capable of writing code that can call itself to improve itself. We consider concerns around the development of self-improving technologies and evaluate the frequency with which the generated code bypasses a sandbox.
Large language models (LLMs) have revolutionized zero-shot task performance, mitigating the need for task-specific annotations while enhancing task generalizability. Despite its advancements, current methods using trigger phrases such as "Let's think step by step" remain limited. This study introduces PRomPTed, an approach that optimizes the zero-shot prompts for individual task instances following an innovative manner of "LLMs in the loop". Our comprehensive evaluation across 13 datasets and 10 task types based on GPT-4 reveals that PRomPTed significantly outperforms both the naive zero-shot approaches and a strong baseline (i.e., "Output Refinement") which refines the task output instead of the input prompt. Our experimental results also confirmed the generalization of this advantage to the relatively weaker GPT-3.5. Even more intriguingly, we found that leveraging GPT-3.5 to rewrite prompts for the stronger GPT-4 not only matches but occasionally exceeds the efficacy of using GPT-4 as the prompt rewriter. Our research thus presents a huge value in not only enhancing zero-shot LLM performance but also potentially enabling supervising LLMs with their weaker counterparts, a capability attracting much interest recently. Finally, our additional experiments confirm the generalization of the advantages to open-source LLMs such as Mistral 7B and Mixtral 8x7B.