5fa1a76
1
2
feed forward chunking In each residual attention block in transformers the self-attention layer is usually followed by 2 feed forward layers.
