# Basic [[Transformer]]
- ![[Pasted image 20220307183126.webp]]
- ![[Pasted image 20220621164717.webp]]
- Feed forward blocks, are two [[Dense]] MLPs with [[Relu]]. Residual connections in between
- Uses [[Attention]]
- [[Embedding]] [[Layers]] transform between 1 hot and vector rep
- [[Position Encoding]] + [[Token Embedding]]
- [[Position Wise Feed Forward]]