Dive deep into the Transformer Neural Network Architecture for language translation in this comprehensive 28-minute video. Explore key concepts including batch data processing, fixed-length sequences, embeddings, positional encodings, query/key/value vectors, masked multi-head self-attention, residual connections, layer normalization, decoder architecture, cross-attention mechanisms, tokenization, and word generation. Gain practical insights through a Transformer inference example and access additional resources for further learning on neural networks, machine learning, and related mathematical concepts.
Overview
Syllabus
Introduction
Transformer at a high level
Why Batch Data? Why Fixed Length Sequence?
Embeddings
Positional Encodings
Query, Key and Value vectors
Masked Multi Head Self Attention
Residual Connections
Layer Normalization
Decoder
Masked Multi Head Cross Attention
Tokenization & Generating the next translated word
Transformer Inference Example
Taught by
CodeEmporium
Related Courses
-
Deep Dive into the Transformer Encoder Architecture
-
Blowing Up Transformer Decoder Architecture
-
Attention Is All You Need - Transformer Paper Explained
-
Decoder Flow in Transformer Model
-
Transformer Models: Understanding Their Architecture and Functionality - Part 3
-
Transformer Neural Networks, ChatGPT's Foundation, Clearly Explained
