Transform model explanation

Table of contents

game is a game

 With Beijing: Winter Olympics

 transform: encode, decode 12 steps

 The self-attention mechanism is the dismantling comparison of Transformers: generate parts V and weight K, and feedforward neural network for weight adjustment: preliminary deformation 

Encoder Attention is to consider context information

Attention Mechanism: Adopt Multi-Attention Mechanism: Prevent One Person’s Mutiny from Leading to Model Failure

 Data flow: Use algorithms to vectorize words, the same bet. 512 bits

 Through weights: Q, K, V to perform calculations to generate component descriptions and relationship descriptionsEdit

 8 weight matrices to prevent failure and eliminate the influence of initial weightsEdit

Word vectorization, matrix multiplication to generate relationship descriptions, attention weights, and final weighted sumsEdit

---

game is a game

 With Beijing: Winter Olympics

 

 

 

 transform: encode, decode 12 steps

 The self-attention mechanism is the dismantling comparison of Transformers: generate parts V and weight K, and feedforward neural network for weight adjustment: preliminary deformation 

Encoder Attention is to consider context information

[Transformer model] Graceful animation is easy to learn, image metaphor is easy to remember_哔哩哔哩_bilibili

Attention Mechanism: Adopt Multi-Attention Mechanism: Prevent One Person’s Mutiny from Leading to Model Failure

 

 Data flow: Use algorithms to vectorize words, the same bet. 512 bits

 

 Calculate and generate component description and relationship description through weight: Q, K, V

 

 8 weight matrices to prevent failure and eliminate the influence of initial weights

 

Word vectorization, matrix multiplication to generate relationship description, attention weight, and final weighted sum