EpiCoder-func-380k is a dataset released by Microsoft in 2025, focusing on fine-tuning large language models for code generation capabilities. The related research paper is titled "EpiCoder: Encompassing Diversity and Complexity in Code Generation", aimed at addressing the challenges of instruction diversity and complexity in code generation tasks.

The dataset contains 380,000 function-level instruction-output paired instances, specifically designed for training large language models to generate Python code based on detailed textual instructions. The dataset is constructed through a synthetic method, where both instructions and code are generated by large language models and validated for correctness in Docker environments, covering various programming scenarios including instruction-level parallel optimization, making it suitable for improving model performance on specific coding tasks.

Dataset Composition

The dataset mainly consists of the following two fields, with each entry representing a complete programming task instance:

• instruction: String type, containing a detailed programming task description that specifies the functionality, constraints, and optimization requirements of the code.
• output: String type, containing the Python code implementation that satisfies the above instruction.

All data consists of expert-generated instructions paired with model-synthesized code, without additional metadata or comment fields.

Citation

@misc{wang2025epicoderencompassingdiversitycomplexity,
      title={EpiCoder: Encompassing Diversity and Complexity in Code Generation}, 
      author={Yaoxiang Wang and Haoling Li and Xin Zhang and Jie Wu and Xiao Liu and Wenxiang Hu and Zhongxin Guo and Yangyu Huang and Ying Xin and Yujiu Yang and Jinsong Su and Qi Chen and Scarlett Li},
      year={2025},
      eprint={2501.04694},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2501.04694}, 
}