CSC 6714 - Large Language Models

• Describe the inference process for a token-prediction transformer model including token generation
• Explain the structure and mechanics of the transformer architecture and its components, including recent advances
• Train a small transformer model from scratch to predict the next token in a sequence
• Compare techniques for training transformer models including pretraining, instruction fine-tuning, and reinforcement learning
• Demonstrate working applications of LLMs on real-world problems of varying difficulty
• Use advanced prompting techniques such as chain of thought and structured output to improve LLM performance on a task
• Evaluate the performance of an LLM on tasks using approaches such as perplexity and AI as judge

• Python programming experience including the use of data science and machine learning libraries
• Able to perform and interpret matrix and vector arithmetic including addition, dot products, and matrix-vector multiplication
• Able to train and apply various classical machine learning models for classification and regression problems
• Able to design and execute experiments to evaluate machine learning models
• Able to interpret metrics such as accuracy, precision, and recall to evaluate model prediction performance

• The modern transformer architecture including embedding components
• Process for inference and token generation from transformer architectures
• Computational efficiency of and optimizations for transformer inference
• Techniques for training LLMs as foundation and instruction-following models
• Examples of LLM applications (varies depending on instructor preference)
• Foundational and advanced prompt engineering techniques such as prompt structure, chain of thought, few-shot learning, and structured output
• Evaluation of LLMs including challenges caused by and approaches for evaluating open-ended output