Building LLM Applications with Prompt Engineering

• Train neural networks across multiple servers
• Use techniques such as activation checkpointing, gradient accumulation, and various forms of model parallelism to overcome the challenges associated with large-model memory footprint
• Capture and understand training performance characteristics to optimize model architecture
• Deploy very large multi-GPU models to production using NVIDIA Triton™ Inference Server

Module 1:  Introduction

• Orient to the main workshop topics, schedule and prerequisites.
• Learn why prompt engineering is core to interacting with Large Language Models (LLMs).
• Discuss how prompt engineering can be used to develop many classes of LLM-based applications.
• Learn about NVIDIA LLM NIM, used to deploy the Llama 3.1 LLM used in the workshop.

Module 2: Introduction to Prompting

• Get familiar with the workshop environment.
• Create and view responses from your first prompts using the OpenAI API, and LangChain.
• Learn how to stream LLM responses, and send LLMs prompts in batches, comparing differences in performance.
• Begin practicing the process of iterative prompt development.
• Create and use your first prompt templates.
• Do a mini project where to perform a combination of analysis and generative tasks on a batch of inputs.

Module 3: LangChain Expression Language (LCEL), Runnables, and Chains   

• Learn about LangChain runnables, and the ability to compose them into chains using LangChain Expression Language (LCEL).
• Write custom functions and convert them into runnables that can be included in LangChain chains.
• Compose multiple LCEL chains into a single larger application chain.
• Exploit opportunities for parallel work by composing parallel LCEL chains.
• Do a mini project where to perform a combination of analysis and generative tasks on a batch of inputs using LCEL and parallel execution.

Module 4: Prompting With Messages

• Learn about two of the core chat message types, human and AI messages, and how to use them explicitly in application code.
• Provide chat models with instructive examples by way of a technique called few-shot prompting.
• Work explicitly with the system message, which will allow you to define an overarching persona and role for your chat models.
• Use chain-of-thought prompting to augment your LLMs ability to perform tasks requiring complex reasoning.
• Manage messages to retain conversation history and enable chatbot functionality.
• Do a mini-project where you build a simple yet flexible chatbot application capable of assuming a variety of roles.

Module 5: Structured Output

• Explore some basic methods for using LLMs to generate structured data in batch for downstream use.
• Generate structured output through a combination of Pydantic classes and LangChain's `JsonOutputParser`.
• Learn how to extract data and tag it as you specify out of long form text.
• Do a mini-project where you use structured data generation techniques to perform data extraction and document tagging on an unstructured text document.

Module 6: Tool Use and Agents

• Create LLM-external functionality called tools, and make your LLM aware of their availability for use.
• Create an agent capable of reasoning about when tool use is appropriate, and integrating the result of tool use into its responses.
• Do a mini-project where you create an LLM agent capable of utilizing external API calls to augment its responses with real-time data.

Module 7: Assessment and Final Review

• Review key learnings and answer questions.
• Earn a certificate of competency for the workshop.
• Complete the workshop survey.
• Get recommendations for the next steps to take in your learning journey.