barty/EvalRepair-Java

1---2license: mit3---4 5## Dataset Description6 7- **Repository:** [MORepair](https://github.com/buaabarty/morepair)8- **Paper:** [MORepair: Teaching LLMs to Repair Code via Multi-Objective Fine-tuning](https://arxiv.org/abs/2404.12636)9- **Point of Contact:** [Boyang Yang](mailto:yby@ieee.org)10 11### Dataset Summary12 13EvalRepair-Java is a benchmark for evaluating Java program repair performance, derived from HumanEval. It contains 163 single-function repair tasks, each with a buggy implementation and its corresponding fixed version.14 15### Supported Tasks16 17- Program Repair: Fixing bugs in Java functions18- Code Generation: Generating correct implementations from buggy code19 20### Dataset Structure21 22Each row contains:23- `task_id`: Unique identifier for the task (same as HumanEval)24- `buggy_code`: The buggy implementation25- `fixed_code`: The correct implementation26- `unit_test`: Unit tests for verifying the correctness of the implementation27 28### Source Data29 30This dataset is derived from HumanEval, a benchmark for evaluating code generation capabilities. We manually introduced bugs into the original implementations and verified the fixes.31 32### Citation33 34```bibtex35@article{morepair,36author = {Yang, Boyang and Tian, Haoye and Ren, Jiadong and Zhang, Hongyu and Klein, Jacques and Bissyande, Tegawende and Le Goues, Claire and Jin, Shunfu},37title = {MORepair: Teaching LLMs to Repair Code via Multi-Objective Fine-Tuning},38year = {2025},39publisher = {Association for Computing Machinery},40issn = {1049-331X},41url = {https://doi.org/10.1145/3735129},42doi = {10.1145/3735129},43journal = {ACM Trans. Softw. Eng. Methodol.},44}45```