LLMOps

What You'll Learn

• Adapt an open-source pipeline to apply supervised fine-tuning for better question answering.
• Implement best practices for versioning data and models, and pre-process large datasets within a data warehouse.
• Apply responsible AI practices by generating safety scores on harmful content sub-categories.

About This Course

This course provides a comprehensive introduction to the LLMOps pipeline, covering the steps to adapt a supervised tuning pipeline to create a custom LLM workflow for specific applications, such as developing a question-answer chatbot for Python coding questions.

• Retrieve and transform training data for supervised fine-tuning of an LLM.
• Version data and tuned models to effectively track tuning experiments.
• Configure and execute an open-source supervised tuning pipeline to train and deploy a tuned LLM.
• Monitor safety scores to responsibly filter and evaluate LLM application behavior.

Hands-on practice includes tools like BigQuery data warehouse, Kubeflow Pipelines, and Google Cloud, to gain proficiency in building and managing an LLMOps pipeline.

Course Outline

• Introduction
  Overview of the LLMOps pipeline and its applications in LLM training and deployment.
• The Fundamentals
  Core principles of LLM fine-tuning, data versioning, and model management.
• Data Preparation
  Techniques for preparing and pre-processing data for supervised instruction tuning, with hands-on examples.
• Automation and Orchestration with Pipelines
  Implementing automated pipelines for training and deploying LLMs with code examples.
• Prediction, Prompts, Safety
  Configuring LLM prompts, predicting outputs, and monitoring safety metrics.
• Conclusion
  Summary of learned concepts and tools for ongoing application in LLMOps.
• Next Step
  Code examples and guidance for applying pipeline techniques to custom projects.

Who Should Join?

This course is ideal for anyone interested in LLM fine-tuning and building LLMOps pipelines. Basic Python knowledge and familiarity with machine learning workflows are recommended.