Please enable JavaScript.
Coggle requires JavaScript to display documents.
Transformer - Coggle Diagram
Transformer
Treinamento
Matrizes de Pesos
Embeeding
Atenção (Wq, Wk, Wv)
Positional Encoding
Não é treinável no Transformer original
Feed Forward (w1 e w2)
Saída (Wo)
Back Propagation (Minimizar erro da saída do modelo)
Tokens (O cachorro late)
Embeddings
Positional Encoding
Attention
Feed Forward
Linear
Softmax
Output
Encoder (Self Attention)
Decoder (Mask Attention)