Deep Learning Explainer
This video explains the Transformer architecture in a very detailed way, including most math formulas in the paper, and the neural network operations behind it. The Transformer is the foundation of many powerful language models like BERT, GPT3, RoBERTa, XLNET, ELECTRA, T5. Understanding how it works in detail might help you modify, optimize, or improve it in the way you want.
0:00 – Intro
1:10 – Architecture overview
1:56 – Encoder
3:58 – Residual connection & layer normalization
6:59 – Decoder
11:14 – Attention mechanism
14:30 – Scaled dot-product attention
20:22 – Learned projection layers
26:32 – Multi-head attention
28-39 – Encoder-decoder attention
31:18 – Encoder self-attention
31:34 – Decoder self-attention
33:58 – Position-wise feedforward network
36:41 – Word embedding
39:34 – Positional encoding
47:37 – Why self-attention
What Is GPT-3 Series
https://www.youtube.com/playlist?list=PLoS8jSwcU-c_j5zDl49skiP0u63dd-C4Q
Paper: Attention Is All You Need
https://arxiv.org/abs/1706.03762
Abstract
The dominant sequence transduction models are based on complex recurrent or convolutional neural networks that include an encoder and a decoder. The best
performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English- to-German translation task, improving over the existing best results, including ensembles, by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.0 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.