Pathway 2: "I Want to Fine-Tune and Train Models" (ML Engineer)

Target audience: ML engineers with existing deep learning experience who want to specialize in LLM training

Goal: Master the full training pipeline from data preparation through alignment, with practical experience on real hardware.

Chapter Guide

• Start Ch 03: Sequence Models and Attention (start here if you need a transformer refresher; otherwise start at Ch 06) refresh attention if needed before diving in
• Focus Ch 06: Pre-training and Scaling Laws core theory: data, compute, and loss scaling
• Focus Ch 07: The Modern LLM Landscape know the architecture landscape you will train
• Skim Ch 08: Reasoning Models and Test-Time Compute understand how reasoning emerges from training
• Focus Ch 09: Inference Optimization optimize inference for the models you build
• Focus Ch 13: Synthetic Data Generation generate high-quality training data at scale
• Focus Ch 14: Fine-Tuning Fundamentals your primary workflow: full fine-tuning
• Focus Ch 15: PEFT (LoRA, QLoRA) parameter-efficient methods for constrained hardware
• Focus Ch 16: Knowledge Distillation and Model Merging compress and combine models efficiently
• Focus Ch 17: Alignment (RLHF, DPO) align models to human preferences
• Skim Ch 18: Interpretability peek inside trained models to debug behavior
• Skim Ch 10: Working with LLM APIs understand how users consume the models you train
• Skim Ch 12: Hybrid ML+LLM Architectures patterns for combining trained models with LLM APIs
• Focus Ch 29: Evaluation and Experiment Design rigorous evaluation of your trained models
• Skim Ch 30: Observability and Monitoring monitor model performance after deployment
• Skim Ch 31: Production Engineering deploy and serve your fine-tuned models
• Skim Ch 34: Emerging Architectures next-generation architectures to watch
• Skim Ch 35: AI and Society safety alignment and open problems in the field