Compact and Domain-Specific LLMs

This unit examines the trade-offs between large general-purpose models and smaller, task-focused models, explores successful examples of compact and domain-specific models, and introduces prompting and tuning strategies. It builds on foundational concepts from earlier units and prepares learners for hands-on tuning and deployment techniques in the final unit.

Learning Objectives

• Analyze the trade-offs (compute, latency, accuracy, cost, and privacy) between large general-purpose models and compact/nimble models in real-world use cases
• Evaluate scenarios and select appropriate model types (general-purpose, compact, or domain-specific) based on technical constraints and application requirements
• Apply prompt engineering techniques with general-purpose models to achieve domain-relevant outputs and compare results to those from smaller models
• Design a basic tuning strategy (e.g., fine-tuning, instruction tuning, or parameter-efficient tuning) for a domain-specific model and justify the chosen approach given target performance and resource constraints