This paper describes Netmarble's submission to WMT21 Automatic Post-Editing\n(APE) Shared Task for the English-German language pair. First, we propose a\nCurriculum Training Strategy in training stages. Facebook Fair's WMT19 news\ntranslation model was chosen to engage the large and powerful pre-trained\nneural networks. Then, we post-train the translation model with different\nlevels of data at each training stages. As the training stages go on, we make\nthe system learn to solve multiple tasks by adding extra information at\ndifferent training stages gradually. We also show a way to utilize the\nadditional data in large volume for APE tasks. For further improvement, we\napply Multi-Task Learning Strategy with the Dynamic Weight Average during the\nfine-tuning stage. To fine-tune the APE corpus with limited data, we add some\nrelated subtasks to learn a unified representation. Finally, for better\nperformance, we leverage external translations as augmented machine translation\n(MT) during the post-training and fine-tuning. As experimental results show,\nour APE system significantly improves the translations of provided MT results\nby -2.848 and +3.74 on the development dataset in terms of TER and BLEU,\nrespectively. It also demonstrates its effectiveness on the test dataset with\nhigher quality than the development dataset.\n