From Data Processing to Fine-Tuning an LLM

This hands-on workshop provides a comprehensive foundation in machine learning fundamentals and practical model fine-tuning. Participants will begin with the core concepts underpinning modern vs classical ML workflow, data preprocessing (outlier detection, class imbalance, feature engineering), and data partitioning strategies. From there, the workshop covers the full training pipeline: loss functions, optimization methods, bias-variance tradeoff, regularization, and evaluation metrics.

The second half transitions to large language models and parameter-efficient fine-tuning. Participants will learn how transformer architectures work, why pre-training enables rapid specialization, and how LoRA (Low-Rank Adaptation) makes it possible to fine-tune a model by training less than 0.3% of its parameters. Attendees will walk through the complete fine-tuning pipeline in code: prompt formatting, tokenization, 4-bit quantization, training configuration, early stopping, and adapter saving.

By the end of the session, participants will have trained and evaluated their own LoRA adapter and will have the tools to apply the same workflow to their own datasets and models.

Prerequisites: Basic Python familiarity. No prior deep learning experience required.

Format: Lecture + live coding. Please bring a laptop.