Hands on session for the ACM Winter School focused on Generative AI

Explore the Hugging Face Hub

• Models
• Datasets
• Spaces
• Tasks
• Chat

Interesting demos:

1. https://huggingface.co/spaces/Nexusflow/NexusRaven-V2-Demo
2. https://huggingface.co/spaces/yunyangx/EfficientSAM
3. https://huggingface.co/spaces/openai/whisper
4. https://huggingface.co/spaces/diffusers/unofficial-SDXL-Turbo-i2i-t2i
5. https://huggingface.co/blog/personal-copilot

Module 1: Prompt Engineering

Set up: Google Colab: https://colab.research.google.com/ - 16GB T4 GPU.

• We start by covering the fundamentals of Prompting. We utilize the mistralai/Mistral-7B-Instruct-v0.2 open access model, which excels at following instructions. We demonstrate how to enable token streaming and employ chat templates.
• We dive into powerful prompting techniques like few-shot, chain-of-thought (CoT), Self Consistency, ReACT (Reason & Act), and Tree of Thoughts (ToT) with coding examples. We'll showcase a Breadth First Search based ToT example and Langchain tools for ReACT example.
• Explore JSON-only outputs, prompt chaining, and LLMs for evaluation. Plus, discover how to create a Gradio chatbot 🚀.

Module 2: Question Answering Chatbot using Retrieval Augmented Generation (RAG)

• We will see how Langchain makes it easy as a framework to bring together most of the steps aboves in simple APIs. We will be using open source models from 🤗 Hub, Langchain and Chroma vector DB.

POC: https://huggingface.co/spaces/smangrul/PEFT-Docs-QA-Chatbot