💡 The Transformer, a purely attention-based architecture without recurrence or convolutions, achieves state-of-the-art performance on translation and parsing with lower training cost, laying the foundat