| Implement those changes which often means changing the self-attention layer, the order of the normalization | |
| layer, etc… Again, it is often useful to look at the similar architecture of already existing models in Transformers to | |
| get a better feeling of how your model should be implemented. |