Live Python Demos - Generative Artificial Intelligence

Overview

Explore a comprehensive 1-hour 27-minute live Python demonstration focused on Generative Artificial Intelligence. Dive into the world of hosting Large Language Models (LLMs) and Large Multimodal Models (LMMs) on personal laptops, specifically a MacBook Pro m3 max with 48GB RAM. Witness real-time image processing and prompt responses, with the server achieving speeds of 130+ tokens/second. Discover the utilization of Hugging Chat platform, line-by-line coding with Hugging Face's Blip and Gradio UI, and a LangChain agent executor demo evaluated using LangSmith. Learn about cost determination through OpenAI's Platform and Colab. Explore the setup of Open WebUI with Ollama, Docker, and Portainer for local access to open-source generative AI models. Compare the output quality and speed of multiple LLMs and multimodal models, including moondream 1.8b, llava 7b, and llava 13b. Examine code generation tasks using deepseek-coder-v2:16b and codestral 22.2B, with a focus on Hugging Face's pipeline for high-quality photorealistic image generation. Witness a live demonstration of cross-device functionality as an iPhone 15 Pro successfully interacts with the laptop server to generate accurate responses to image prompts.

Syllabus

deepseek-coder-vb and codestral 22.2B were asked to produce code regarding three different python generative ai tasks. Responses typically favored code featuring older models, however deepseek-coder’s response to use Hugging Face’s pipeline to generate high quality photorealistic images is still relevant today. Lastly, the WebUI application was accessed by a second user account on an iPhone 15 Pro on a different cellular network live to successfully send an image of Stanford Medicine students standing in a formal photograph with the prompt “Describe” to the laptop server, and the phone received an accurate generated response.