Skywork series models are pre-trained on 3.2TB of high-quality multilingual (mainly Chinese and English) and code data. We have open-sourced the model, training data, evaluation data, evaluation methods, etc. 天工系列模型在3.2TB高质量多语言和代码数据上进行预训练。我们开源了模型参数，训练数据，评估数据，评估方法。
你好，我看git和huggingface上面的MOCK_GSM8K_TEST数据集只有答案部分，想问一下问题是怎么构造的，可以开源出来吗