QANet is a novel architecture for machine reading comprehension and question answering (Q&A) that eliminates the need for recurrent neural networks (RNNs). Instead, it uses convolution and self-attention mechanisms to model local and global interactions, respectively. This approach significantly reduces training and inference time, making it 3x to 13x faster in training and 4x to 9x faster in inference compared to RNN-based models. The model achieves competitive accuracy on the SQuAD dataset, outperforming existing models with up to 13x speedup. Additionally, QANet benefits from data augmentation techniques, such as back-translation, which enriches the training data and further improves performance. On the SQuAD dataset, QANet achieves an F1 score of 84.6, surpassing the best published result of 81.8. The paper also includes ablation studies and robustness experiments to validate the effectiveness of the proposed model.QANet is a novel architecture for machine reading comprehension and question answering (Q&A) that eliminates the need for recurrent neural networks (RNNs). Instead, it uses convolution and self-attention mechanisms to model local and global interactions, respectively. This approach significantly reduces training and inference time, making it 3x to 13x faster in training and 4x to 9x faster in inference compared to RNN-based models. The model achieves competitive accuracy on the SQuAD dataset, outperforming existing models with up to 13x speedup. Additionally, QANet benefits from data augmentation techniques, such as back-translation, which enriches the training data and further improves performance. On the SQuAD dataset, QANet achieves an F1 score of 84.6, surpassing the best published result of 81.8. The paper also includes ablation studies and robustness experiments to validate the effectiveness of the proposed model.