Why can transformers train faster on long text than RNNs?
- Attention sees all positions at once
- They use far fewer parameters overall
- They need no training data
- They read text right to left
RNNs process tokens one step at a time, while self-attention handles every position in parallel.
Play AI & ML ▸