Large Language Models Bootcamp

Unleash your creativity and efficiency with prompt engineering. Seamlessly prompt models, control outputs, and generate captivating content across various domains and tasks.

• Prompt Design and Engineering
  • Crafting Instructions for Effective Prompting
  • Utilizing Examples to Guide Model Behavior
• Managing Model Outputs
  • Determining Optimal Stopping Points
  • Balancing Creativity and Predictability
  • Strategies for Saving and Sharing Prompts
• Innovative Use Case Development
  • Tailoring Prompts to Goals, Tasks, and Domains
  • Practical Examples:
    • Summarizing Complex Reports
    • Extracting Sentiment and Key Topics from Texts
• Understanding and Mitigating Prompt Engineering Risks
  • Identifying Common Risks: Prompt Injection, Prompt Leaking, Jailbreaking
  • Best Practices for Secure Prompt Engineering
• Advanced Prompting Techniques
  • Enhancing Performance with Few-Shot and Chain-of-Thought (CoT) Prompting
  • Exploring Program-aided Language Models (PAL) and ReAct Methods

In-depth discussion on fine-tuning of large language models through theoretical discussions, exploring rationale, limitations, and Parameter Efficient Fine Tuning.

• Fine Tuning Foundation LLMs
  • Transfer learning and Fine-tuning
  • Different fine-tuning techniques
  • Limitations for fine-tuning
  • Parameter-efficient fine-tuning in depth.
    * Quantization of LLMs
    * Low-Rank Adaptation (LoRA) and QLoRA
  • Fine-tuning vs. RAG: When to use one or the other. Risks and limitations.
• Hands-on Exercise:
  • In-Class: Instruction fine-tuning, deploying, and evaluating a LLaMA2-7B 4-bit quantized model
  • Homework: Fine-tuning and deploying OpenAI and Llama models on Azure AI studio.

Build LLM Apps using LangChain. Learn about LangChain's key components such as models, prompts, parsers, memory, chains, and Question-Answering.

• Introduction to LangChain:
  • Why do we need an orchestration tool for LLM application development?
  • What is LangChain?
  • Different components of LangChain
• Why are Orchestration Frameworks Needed?
  • Eliminate the need for foundation model retraining
  • Overcoming token limits
  • Connecters for data sources.
• Interface with any LLM using Model I/O
  • Model I/O overview
  • Components of Model I/O: prompts, language models, and output parsers.
  • Overview of prompts, prompt templates, and example selectors.
  • Different types of models: language, chat, and embedding models
  • Structuring language model responses using various types of output parsers
• Connecting external data with LLM application with retrieval
  • Retrieval overview
  • The rationale for the requirement of retrieval and how does it work with LangChain
  • Components of retrieval: document loaders, transformer, embedding, vector store and retrieval
  • Loading public, private, structured, and unstructured data with document loaders
  • Transforming documents to fewer chunks and extracting metadata using document transformers
  • Embedding and vector stores for converting documents into vectors and for efficient storage and retrieval.
  • Optimizing retrieval using different retrieval techniques available in LangChain
• Creating complex LLM workflows with Chains
  • Chains overview
  • Various foundational chain types: LLM, Router, Sequential, and Transformation
  • Summarizing large documents using different document chains like stuff, refine, map reduce.
• Retain Context and Refer to Past Interactions with the memory component
  • How memory can empower AI applications
  • Different types of memories: buffer, buffer window, token buffer memory, entity memory, knowledge graph, summary, vector store
  • Overcoming token limit by using memory based on summarization of past conversations
  • Knowledge graph-based memory
  • Vector backed memory
• Dynamic Decision-Making with LLMs Using Agents
  • Agent