Hate speech detection on social media networks is the classification task that automatically detects harmful comments from users and prevents the appearance of those toxic comments on social sites. The profit of the hate speech detection task is preventing harassment and toxicity content on the social networks site to protect the users that join the social media networks. However, hate speech detection is challenging, especially for low-resource languages like Vietnamese. In this paper, we investigate the effect of the transformer-based language model and data augmentation techniques for hate speech detection on Vietnamese texts. Then, we proposed our ensemble model to boost the accuracy of hate speech detection in Vietnamese texts.