In the recent past, social media platforms have helped people in connecting\nand communicating to a wider audience. But this has also led to a drastic\nincrease in cyberbullying. It is essential to detect and curb hate speech to\nkeep the sanity of social media platforms. Also, code mixed text containing\nmore than one language is frequently used on these platforms. We, therefore,\npropose automated techniques for hate speech detection in code mixed text from\nscraped Twitter. We specifically focus on code mixed English-Hindi text and\ntransformer-based approaches. While regular approaches analyze the text\nindependently, we also make use of content text in the form of parent tweets.\nWe try to evaluate the performances of multilingual BERT and Indic-BERT in\nsingle-encoder and dual-encoder settings. The first approach is to concatenate\nthe target text and context text using a separator token and get a single\nrepresentation from the BERT model. The second approach encodes the two texts\nindependently using a dual BERT encoder and the corresponding representations\nare averaged. We show that the dual-encoder approach using independent\nrepresentations yields better performance. We also employ simple ensemble\nmethods to further improve the performance. Using these methods we report the\nbest F1 score of 73.07% on the HASOC 2021 ICHCL code mixed data set.\n