RobuNFR: Evaluating the Robustness of Large Language Models on Non-Functional Requirements Aware Code Generation

When using LLMs to address Non-Functional Requirements (NFRs), developers may behave differently (e.g., expressing the same NFR in different words). Robust LLMs should output consistent results across these variations; however, this aspect remains underexplored. We propose RobuNFR for evaluating the robustness of LLMs in NFR-aware code generation across four NFR dimensions: design, readability, reliability, and performance, using three methodologies: prompt variation, regression testing, and diverse workflows. Our experiments show that RobuNFR reveals robustness issues in the tested LLMs when considering NFRs in code generation. Specifically, under prompt variation, including NFRs leads to a decrease in Pass@1 by up to 39 percent and an increase in the standard deviation from 0.48 to 2.48 compared to the baseline without NFRs (i.e., Function-Only). While incorporating NFRs generally improves overall NFR metrics, it also results in higher prompt sensitivity. In regression settings, some LLMs exhibit differences across versions, with improvements in one aspect (e.g., reduced code smells) often accompanied by regressions in another (e.g., decreased correctness), revealing inconsistencies that challenge their robustness. When varying workflows, the tested LLMs show significantly different NFR-aware code generation capabilities between two workflows: (1) integrating NFRs and functional requirements into the initial prompt and (2) enhancing Function-Only-generated code with the same NFR.

When LLM-based Code Generation Meets the Software Development Process

Software process models play a pivotal role in fostering collaboration and communication within software teams, enabling them to tackle intricate development tasks effectively. This paper introduces LCG, a code generation framework inspired by established software engineering practices. LCG leverages multiple Large Language Model (LLM) agents to emulate various software process models, namely LCGWaterfall, LCGTDD, and LCGScrum. Each model assigns LLM agents specific roles such as requirement engineer, architect, developer, tester, and scrum master, mirroring typical development activities and communication patterns. Through collaborative efforts utilizing chain-of-thought and prompt composition techniques, the agents continuously refine themselves to enhance code quality. Utilizing GPT3.5 as the underlying LLM and baseline (GPT), we evaluate LCG across four code generation benchmarks: HumanEval, HumanEval-ET, MBPP, and MBPP-ET. Results indicate LCGScrum outperforms other models, achieving Pass@1 scores of 75.2, 65.5, 82.5, and 56.7 in HumanEval, HumanEval-ET, MBPP, and MBPP-ET, respectively - an average 15% improvement over GPT. Analysis reveals distinct impacts of development activities on generated code, with design and code reviews contributing to enhanced exception handling, while design, testing, and code reviews mitigate code smells. Furthermore, temperature values exhibit negligible influence on Pass@1 across all models. However, variations in Pass@1 are notable for different GPT3.5 model versions, ranging from 5 to over 60 in HumanEval, highlighting the stability of LCG across model versions. This stability underscores the importance of adopting software process models to bolster the quality and consistency of LLM-generated code.