HumanEvo: An Evolution-Aware Benchmark for More Realistic Evaluation of Repository-Level Code Generation
Open accessAbstract
To evaluate the repository-level code generation capabilities of Large Language Models (LLMs) in complex real-world software development scenarios, many evaluation methods have been developed. These methods typically leverage contextual code from the latest version of a project to assist LLMs in accurately generating the desired function. However, such evaluation methods fail to consider the dynamic evolution of software projects over time, which we refer to as evolution-ignored settings. This in turn results in inaccurate evaluation of LLMs' performance. In this paper, we conduct an empirical study to deeply understand LLMs' code generation performance within settings that reflect the evolution nature of software development. To achieve this, we first construct an evolution-aware repository-level code generation dataset, namely HumanEvo, equipped with an automated execution-based evaluation tool. Second, we manually categorize HumanEvo according to dependency levels to more comprehensively analyze the model's performance in generating functions with different dependency levels. Third, we conduct extensive experiments on HumanEvo with seven representative and diverse LLMs to verify the effectiveness of the proposed benchmark. We obtain several important findings through our experimental study. For example, we find that previous evolution-ignored evaluation methods result in inflated performance of LLMs, with performance overestimations ranging from 10.0% to 61.1% under different context acquisition methods, compared to the evolution-aware evaluation approach. Based on the findings, we give actionable suggestions for more realistic evaluation of LLMs on code generation. We also build a shared evolution-aware code generation toolbox to facilitate future research. The replication package including source code and datasets is anonymously available at https://github.Com/DeepSoftwareAnalytics/HumanEvo.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
Competition-level code generation with AlphaCode
Li, Y., Choi, D., Chung, J., Kushman, N., Schrittwieser, J., Leblond, R., Eccles, T., Keeling, J., Gimeno, F., Dal Lago, A., et al. (2022) · Science · doi:10.1126/science.abq1158 · Cited by 1,043 · see citing works
ChatDev: Communicative Agents for Software Development
Qian, C., Liu, W., Liu, H., Chen, N., Dang, Y., Li, J., Yang, C., Chen, W., Su, Y., Cong, X., et al. (2024) · Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2024.acl-long.810 · Cited by 490 · see citing works
CodeGeeX: A Pre-Trained Model for Code Generation with Multilingual Benchmarking on HumanEval-X
Zheng, Q., Xia, X., Zou, X., Dong, Y., Wang, S., Xue, Y., Shen, L., Wang, Z., Wang, A., Li, Y., et al. (2023) · Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining · doi:10.1145/3580305.3599790 · Cited by 263 · cited 1× · see citing works
“Many code LLMs [8]–[10], [17], [18], [29]–[33], [39], [40], [51] are being employed to empower programming assistants, which are playing a vital role in practical * Yanlin Wang is the corresponding author. software development [70]–[73].”
Self-collaboration code generation via ChatGPT
Dong, Y., Jiang, X., Jin, Z., & Li, G. (2024) · ACM Transactions on Software Engineering and Methodology · doi:10.1145/3672459 · Cited by 256 · cited 1× · see citing works
“In recent years, the LLM-based code generation task has drawn widespread attention [24], [25], [36]–[38], [52]–[54], [75], [76].”
RepoCoder: Repository-Level Code Completion Through Iterative Retrieval and Generation
Zhang, F., Chen, B., Zhang, Y., Keung, J., Liu, J., Zan, D., Mao, Y., Lou, J.-G., & Chen, W. (2023) · Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2023.emnlp-main.151 · Cited by 173 · see citing works
CoderEval: A Benchmark of Pragmatic Code Generation with Generative Pre-trained Models
Yu, H., Shen, B., Ran, D., Zhang, J., Zhang, Q., Ma, Y., Liang, G., Li, Y., Wang, Q., & Xie, T. (2024) · Proceedings of the IEEE/ACM 46th International Conference on Software Engineering · doi:10.1145/3597503.3623316 · Cited by 140 · see citing works
Self-citationZhang, K., Li, J., Li, G., Shi, X., & Jin, Z. (2024) · Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2024.acl-long.737 · Cited by 138 · see citing works
CodePlan: Repository-level Coding using LLMs and Planning
Bairi, R., Sonwane, A., Kanade, A., C, V. D., Iyer, A., Parthasarathy, S., Rajamani, S., Ashok, B., & Shet, S. (2023) · arXiv · doi:10.48550/arxiv.2309.12499 · Cited by 92 · see citing works
CodePlan: Repository-Level Coding using LLMs and Planning
Bairi, R., Sonwane, A., Kanade, A., C., V. D., Iyer, A., Parthasarathy, S., Rajamani, S., Ashok, B., & Shet, S. (2024) · Proceedings of the ACM on Software Engineering · doi:10.1145/3643757 · Cited by 92 · see citing works
Teaching Code LLMs to Use Autocompletion Tools in Repository-Level Code Generation
Wang, C., Zhang, J., Feng, Y., Li, T., Sun, W., Liu, Y., & Peng, X. (2025) · ACM Transactions on Software Engineering and Methodology · doi:10.1145/3714462 · Cited by 40 · cited 1× · see citing works
“Many technical studies have been proposed to better tackle repository-level code generation tasks [2], [4], [6], [48]–[50], [65]–[69], [77], [78].”
BioCoder: a benchmark for bioinformatics code generation with large language models
Tang, X., Qian, B., Gao, R., Chen, J., Chen, X., & Gerstein, M. B. (2024) · Bioinformatics · doi:10.1093/bioinformatics/btae230 · Cited by 30 · see citing works
Flagged first, then by citations
- Authors
- Zheng, D., Wang, Y., Shi, E., Zhang, R., Ma, Y., Zhang, H., & Zheng, Z.
- Affiliations
- Sun Yat-sen University (CN)
- Huawei Technologies (China) (CN)
- Chongqing University (CN)
- Venue
- 2025 IEEE/ACM 47th International Conference on Software Engineering (ICSE)
- Publisher
- IEEE
- Published
- 2025-04-26
- Pages
- 1372-1384
- ISSN
- 0270-5257 (ISSN-L)
- OpenAlex
- W4411552165
- Cited by
- 11
- References
- 11
- Citations by year
- 2026: 4 · 2025: 6 · 2024: 1
- FWCI
- 9.18
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- ACM/IEEE International Conference on Software Engineering: Companion Proceedings (ICSE-Companion) (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Natural Science Foundation of China (62032025)
- Topics
- Physical Sciences › Computer Science › Information Systems › Software Engineering Research
- Physical Sciences › Computer Science › Information Systems › Software Engineering Techniques and Practices
- Physical Sciences › Computer Science › Software › Software Testing and Debugging Techniques
- Keywords
- repository-level code generation, large language models, software evolution, function generation, dependency level, context acquisition
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
