MIGRATION-BENCH
收藏资源简介:
MIGRATION-BENCH是一个大规模的基准数据集,用于评估大型语言模型(LLMs)从Java 8迁移到最新长期支持(LTS)版本(Java 17, 21)的代码迁移能力。该数据集包括一个完整的数据集和其分别包含5,102和300个存储库的子集。选定的子集是为复杂性和难度而精心策划的,为代码迁移领域的研究提供了一个多功能的资源。此外,我们还提供了一个全面的评估框架,以促进对LLMs在此项具有挑战性任务上的严格和标准化的评估。我们进一步提出了SD-Feedback并证明了LLMs可以有效处理到Java 17的存储库级代码迁移。对于包含Claude-3.5-Sonnet-v2的选定子集,SD-Feedback分别实现了62.33%和27.00%的成功率(pass@1),用于最小和最大迁移。基准数据集和源代码可以在https://huggingface.co/collections/AmazonScience和https://github.com/amazon-science/self_debug上获得。
MIGRATION-BENCH is a large-scale benchmark dataset for evaluating the code migration capability of Large Language Models (LLMs) when migrating from Java 8 to the latest long-term support (LTS) versions (Java 17, 21). This dataset includes a full dataset and its two subsets consisting of 5,102 and 300 repositories respectively. The selected subsets are carefully curated for complexity and difficulty, serving as a versatile resource for research in the code migration field. In addition, we provide a comprehensive evaluation framework to facilitate rigorous and standardized assessment of LLMs on this challenging task. We further propose SD-Feedback and demonstrate that LLMs can effectively handle repository-level code migration to Java 17. For the selected subset containing Claude-3.5-Sonnet-v2, SD-Feedback achieves success rates (pass@1) of 62.33% and 27.00% for minimal and maximal migration tasks respectively. The benchmark dataset and source code are available at https://huggingface.co/collections/AmazonScience and https://github.com/amazon-science/self_debug.

- 1MIGRATION-BENCH: Repository-Level Code Migration Benchmark from Java 8AWS AI Labs · 2025年



