Omartificial-Intelligence-Space/Arabic_Reasoning_Dataset

该数据集包含9.21K行阿拉伯语指令推理问答对，旨在增强阿拉伯语模型的推理能力。数据集由三个来源组成：Hugging Face数据集、GPT-4o-Mini API生成的合成数据以及另一个Hugging Face数据集经过过滤和组合的数据。数据集包含两个主要列：instruction（指令或问题）和answer（答案或推理过程）。处理步骤包括数据组合、格式统一和清理。预期用途是训练和微调阿拉伯语大语言模型，并评估其推理和指令遵循能力。注意事项包括数据集的合成性质、覆盖范围和质量的局限性。

This dataset contains 9.21K rows of Arabic instruction-based reasoning QA pairs, designed to enhance the reasoning capabilities of models in Arabic. The dataset is generated by combining original data, synthetic data, and handcrafted data, ensuring consistent formatting and cleaning to maintain data quality. It is intended for training and fine-tuning large Arabic language models, as well as evaluating reasoning and instruction-following performance in Arabic.