(Image-GPT) Generative Pretraining from Pixels | Paper Explained + Colab Notebook

More videos from the GPT series are on my channel. In this paper, OpenAi replaced the sequence of words with pixels and they achieved interesting image completions and generations. So in this video, I will explain concepts like: the motivation behind feeding the GPT with sequence of pixels context reduction method the results they achieved the idea of fine-tuning and linear probing the limitations of transformer architecture I also prepared the intuitive Colab notebook with the weights released by OpenAi! So you can test the capabilities of the model by yourself!   Enjoy!  ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ Paper: https://cdn.openai.com/papers/Generat... ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ Notebook: https://github.com/maciejbalawejder/i... ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ Connect with me on: Linkedin -   / maciej-balawejder-rt8015   GitHub - https://github.com/maciejbalawejder Medium -   / maciejbalawejder   Buy Me a Coffee - [https://www.buymeacoffee.com/mbalawejder](https://www.buymeacoffee.com/mbalawejder) ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ Timestamps: 00:00 Introduction 01:30 Progression of GPTs 03:30 Training 03:57 Different model sizes 04:40 Context reduction 05:37 Overview of the approach 06:10 Linear probing and Fine-tuning 08:23 Results 09:01 Conclusions 10:15 Colab notebook 11:06 Cloning the repo 11:30 Loading the model 12:50 Preprocessing the image 14:10 Generating the completions 15:00 Changes to the original repo