An official website of the United States government
Code Large Language Models (Code LLMs) have excelled at tasks like code completion but often miss deeper semantics such as execution effects and dynamic states. This paper aims to bridge the gap between Code LLMs' reliance on static text data and the need for semantic understanding for complex tasks like debugging and program repair. We introduce a novel strategy, monologue reasoning, to train Code LLMs to reason comprehensive semantics, encompassing high-level functional descriptions, local execution effects of individual statements, and overall input/output behavior, thereby linking static code text with dynamic execution states. We begin by collecting PyX, a clean Python corpus of fully executable code samples with functional descriptions and test cases. We propose training Code LLMs not only to write code but also to understand code semantics by reasoning about key properties, constraints, and execution behaviors using natural language, mimicking human verbal debugging, i.e., rubber-duck debugging. This approach led to the development of SemCoder, a Code LLM with only 6.7B parameters, which shows competitive performance with GPT-3.5-turbo on code generation and execution reasoning tasks. SemCoder achieves 79.3% on HumanEval (GPT-3.5-turbo: 76.8%), 63.6% on CRUXEval-I (GPT-3.5-turbo: 50.3%), and 63.9% on CRUXEval-O (GPT-3.5-turbo: 59.0%). We also study the effectiveness of SemCoder's monologue-style execution reasoning compared to concrete scratchpad reasoning, showing that our approach integrates semantics from multiple dimensions more smoothly. Finally, we demonstrate the potential of applying learned semantics to improve Code LLMs' debugging and self-refining capabilities. Our data, code, and models are available at: https://github.com/ARiSE-Lab/SemCoder.
more »
« less
This content will become publicly available on July 16, 2026
Measuring and Improving the Efficiency of Python Code Generated by LLMs Using CoT Prompting and Fine-Tuning
The burgeoning sophistication of Artificial Intelligence (AI) has catalyzed the rapid proliferation of Large Language Models (LLMs) within software development. These models are increasingly employed to automate the generation of functionally correct code, address complex computational problems, and facilitate the debugging of existing software systems. However, LLM-generated code often faces challenges due to inherent inefficiencies, including redundant logical structures, factually inconsistent content (hallucinations), and programming errors. To address this issue, our research rigorously evaluated the computational efficiency of Python code generated by three prominent LLMs: GPT-4o-Mini, GPT-3.5-Turbo, and GPT-4-Turbo. The evaluation metrics encompass execution time, memory utilization, and peak memory consumption, while maintaining the functional correctness of the generated code. Leveraging the EffiBench benchmark datasets within the Google Vertex AI Workbench environment, across a spectrum of machine configurations, the study implemented a consistent seed parameter to ensure experimental reproducibility. Furthermore, we investigated the impact of two distinct optimization strategies: Chain-of-Thought (CoT) prompting and model fine-tuning. Our findings reveal a significant enhancement in efficiency metrics for GPT-4o-Mini and GPT-3.5-Turbo when employing CoT prompting; however, this trend was not observed for GPT-4-Turbo. Based on its promising performance with CoT prompting, we selected the GPT-4o-Mini model for subsequent fine-tuning, aiming to further enhance both its computational efficiency and accuracy. However, contrary to our expectations, fine-tuning the GPT-4o-Mini model led to a discernible degradation in both its accuracy and computational efficiency. In conclusion, this study provides empirical evidence suggesting that the deployment of high-CPU machine configurations, in synergy with the utilization of the GPT-4o-Mini model and CoT prompting techniques, yields demonstrably more efficient and accurate LLM-generated Python code, particularly within computationally intensive application scenarios.
more »
« less
- Award ID(s):
- 2334243
- PAR ID:
- 10621809
- Publisher / Repository:
- IEEE
- Date Published:
- Journal Name:
- IEEE Access
- Volume:
- 13
- ISSN:
- 2169-3536
- Page Range / eLocation ID:
- 119657 to 119681
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Hedges allow speakers to mark utterances as provisional, whether to signal non-prototypicality or “fuzziness”, to indicate a lack of commitment to an utterance, to attribute responsibility for a statement to someone else, to invite input from a partner, or to soften critical feedback in the service of face management needs. Here we focus on hedges in an experimentally parameterized corpus of 63 Roadrunner cartoon narratives spontaneously produced from memory by 21 speakers for co-present addressees, transcribed to text (Galati and Brennan, 2010). We created a gold standard of hedges annotated by human coders (the Roadrunner-Hedge corpus) and compared three LLM-based approaches for hedge detection: fine-tuning BERT, and zero and few-shot prompting with GPT-4o and LLaMA-3. The best-performing approach was a fine-tuned BERT model, followed by few-shot GPT-4o. After an error analysis on the top performing approaches, we used an LLM-in-the-Loop approach to improve the gold standard coding, as well as to highlight cases in which hedges are ambiguous in linguistically interesting ways that will guide future research. This is the first step in our research program to train LLMs to interpret and generate collateral signals appropriately and meaningfully in conversation.more » « less
-
Instruction tuning is a supervised fine-tuning approach that significantly improves the ability of large language models (LLMs) to follow human instructions. For programming tasks, most models are finetuned with costly human-annotated instruction-response pairs or those generated by large, proprietary LLMs, which may not be permitted. We propose SelfCodeAlign, the first fully transparent and permissive pipeline for self-aligning code LLMs without extensive human annotations or distillation. SelfCodeAlign employs the same base model for inference throughout the data generation process. It first extracts diverse coding concepts from high-quality seed snippets to generate new tasks. It then samples multiple responses per task, pairs each with test cases, and validates them in a sandbox environment. Finally, passing examples are selected for instruction tuning. In our primary experiments, we use SelfCodeAlign with CodeQwen1.5-7B to generate a dataset of 74k instruction-response pairs. Finetuning on this dataset leads to a model that achieves a 67.1 pass@1 on HumanEval+, surpassing CodeLlama-70B-Instruct despite being ten times smaller. Across all benchmarks, this finetuned model consistently outperforms the original version trained with OctoPack, the previous state-of-the-art method for instruction tuning without human annotations or distillation. Additionally, we show that SelfCodeAlign is effective across LLMs of various sizes, from 3B to 33B, and that the base models can benefit more from alignment with their own data distribution. We further validate each component’s effectiveness in our pipeline, showing that SelfCodeAlign outperforms both direct distillation from GPT-4o and leading GPT-3.5-based distillation methods, such as OSS-Instruct and Evol-Instruct. SelfCodeAlign has also led to the creation of StarCoder2-Instruct, the first fully transparent, permissively licensed, and self-aligned code LLM that achieves state-of-the-art coding performance. Overall, SelfCodeAlign shows for the first time that a strong instruction-tuned code LLM can result from self-alignment rather than distillation.more » « less
-
Large language models (LLM) are perceived to offer promising potentials for automating security tasks, such as those found in security operation centers (SOCs). As a first step towards evaluating this perceived potential, we investigate the use of LLMs in software pentesting, where the main task is to automatically identify software security vulnerabilities in source code. We hypothesize that an LLM-based AI agent can be improved over time for a specific security task as human operators interact with it. Such improvement can be made, as a first step, by engineering prompts fed to the LLM based on the responses produced, to include relevant contexts and structures so that the model provides more accurate results. Such engineering efforts become sustainable if the prompts that are engineered to produce better results on current tasks, also produce better results on future unknown tasks. To examine this hypothesis, we utilize the OWASP Benchmark Project 1.2 which contains 2,740 hand-crafted source code test cases containing various types of vulnerabilities. We divide the test cases into training and testing data, where we engineer the prompts based on the training data (only), and evaluate the final system on the testing data. We compare the AI agent’s performance on the testing data against the performance of the agent without the prompt engineering. We also compare the AI agent’s results against those from SonarQube, a widely used static code analyzer for security testing. We built and tested multiple versions of the AI agent using different off-the-shelf LLMs – Google’s Gemini-pro, as well as OpenAI’s GPT-3.5-Turbo and GPT-4-Turbo (with both chat completion and assistant APIs). The results show that using LLMs is a viable approach to build an AI agent for software pentesting that can improve through repeated use and prompt engineering.more » « less
-
As large language models (LLMs) show great promise in generating a wide spectrum of educational materials, robust yet cost-effective assessment of the quality and effectiveness of such materials becomes an important challenge. Traditional approaches, including expert-based quality assessment and student-centered evaluation, are resource-consuming, and do not scale efficiently. In this work, we explored the use of pre-existing student learning data as a promising approach to evaluate LLM-generated learning materials. Specifically, we used a dataset where students were completing the program construction challenges by picking the correct answers among human-authored distractors to evaluate the quality of LLM-generated distractors for the same challenges. The dataset included responses from 1,071 students across 22 classes taught from Fall 2017 to Spring 2023. We evaluated five prominent LLMs (OpenAI-o1, GPT-4, GPT-4o, GPT-4o-mini, and Llama-3.1-8b) across three different prompts to see which combinations result in more effective distractors, i.e., those that are plausible (often picked by students), and potentially based on common misconceptions. Our results suggest that GPT-4o was the most effective model, matching close to 50% of the functional distractors originally authored by humans. At the same time, all of the evaluated LLMs generated many novel distractors, i.e., those that did not match the pre-existing human-authored ones. Our preliminary analysis shows that those appear to be promising. Establishing their effectiveness in real-world classroom settings is left for future work.more » « less
