DUTAOZHANG/Style2Code_datasets

本数据集是基于iamtarun的python_code_instructions_18k_alpacadataset构建的，包含约18000个Python代码片段及其对应的指令。它被设计用于提供指令驱动的代码生成任务的高质量样本。数据集通过使用DeepSeek、Qwen和Doubao三种大型语言模型生成了风格多样的代码样本，并通过后处理清洁和校准，确保样本在语义上等价而风格不同，用于训练对比风格编码器和风格控制生成器。

The dataset is built upon iamtaruns python_code_instructions_18k_alpacadataset, containing approximately 18,000 Python code snippets paired with corresponding instructions. It is designed to provide high-quality samples for instruction-driven code generation tasks. The dataset utilizes three large language models—DeepSeek, Qwen, and Doubao—to generate stylistically diverse code samples, which are then cleaned and aligned post-process to ensure semantic equivalence with distinct styles, used for training a contrastive style encoder and style-controlled generator.