MLOps for LLMs: How To Build, Tune, and Test a Chatbot Without Hating Your Life
Thursday, 16. May 2024 | 08:30 - 17:00
Description
When you work with large machine learning models (like LLMs), the development process can get messy, fast. This workshop will teach you MLOps practices and tools to keep things reproducible, automatable, and testable. We’ll start with the basics of building a chatbot with LLMs, and then show you how to take a prototype and iteratively adapt it to a real-world scenario with software like data version control (DVC). By the end, you’ll have a working pipeline that connects LLMs, a custom dataset, and crucial performance metrics.
Topics
What is MLOps and why do we need it when developing LLMs?
How chatbots use LLMs and retrieval-augmented generation (RAG)
Storing documents in vector embeddings with langchain and faiss
Tracking documents and embeddings with DVC
Retrieving relevant documents and answering questions using an LLM
Chaining the pipeline together with DVC
Measuring performance with ragas
Tracking and comparing performance with DVC
Improving performance by tuning the data, model, or retrieval
Maintaining and updating the application
Requirements
Basic coding skills in Python
Basic familiarity with Git and working in the command line
Target audience
Anyone who wants to learn how to practically build an AI application like a chatbot, and how to incorporate MLOps tools into its development. No previous AI or ML experience is required.