Small and Domain-Specific Models

This unit examines the roles, strengths, and trade-offs of compact and domain-specific language models compared to large general-purpose models. Students will explore motivating examples, real-world success stories, effective prompting techniques, and be introduced to basic tuning strategies to inform model selection and deployment decisions.

Learning Objectives

• Analyze differences and trade-offs between large and compact language models with respect to compute cost, latency, accuracy, privacy, and deployment constraints
• Evaluate case studies of successful smaller and domain-specific models to determine scenarios where they outperform or complement general-purpose models
• Demonstrate effective prompting techniques with general-purpose models to achieve domain-specific outcomes and compare results against using compact models
• Apply selection criteria to propose an initial tuning or deployment strategy (e.g., fine-tuning, parameter-efficient tuning, retrieval augmentation) for a specified use case, justifying the choice based on requirements and constraints