Build and run MiniGPT-4: Downfall of GPT-4?

Build and run MiniGPT-4. MiniGPT-4 is a remarkable model that understands images and communicate with a human on the topic. In this video you will see how to run this model yourself. You'd need about 16GB RAM and 16GB GPU in order to run the 7B version of this model. Github (notebook) link: https://github.com/thushv89/tutorials_deeplearninghero/blob/master/llms/minigpt4.ipynb MiniGPT-4 Paper: https://arxiv.org/pdf/2304.10592.pdf MiniGPT-4 Repo: https://github.com/Vision-CAIR/MiniGPT-4 BLIP-2 Paper: https://arxiv.org/pdf/2301.12597.pdf 00:00 - Introduction 00:52 - Introducing BLIP-2 (Base of MiniGPT-4) 05:13 - Introducing MiniGPT4 06:07 - MiniGPT-4 versions and compute resources 07:13 - Set up the environment 07:43 - Downloading LLMs 08:27 - Caveat: Vicuna versions (v0 vs v1.1) 09:00 - Apply Vicuna deltas 09:27 - Download MiniGPT4 checkpoint 09:46 - Creating MiniGPT4 Chat object 10:23 - Inference with MiniGPT-4 12:01 - Conclusion