This project is a detailed benchmark comparison of state-of-the-art transformer networks namely - BERT, RoBERTa, DistilBERT, ALBERT, XLNet, ERNIE, Longformer and ELECTRA for detection of sexual harassment on Twitter data. Also benchmarked the same for Sarcastic news headlines classification where RoBERTa outperformed all the architectures with accuracies of 83.63% and 90% on the respective datasets. A detailed study was carried out on the comparative performances during the training process and overall evaluation was conducted and documented.