Microsoft · AI-300
Fine-tuning
AI-300 · Lesson 15 of 15 · 1:43:36
YouTube loads only after you press play. This chapter opens at its timestamp in the full course. Watch on YouTube ↗
In this lesson
Prepare for the AI-300 exam (Microsoft Certified: Machine Learning Operations Engineer Associate). In this lesson: supervised fine-tuning, direct preference optimization, reinforcement fine-tuning, and distillation, synthetic training data, hyperparameters, training types, and loss curves, and deploying and evaluating a fine-tuned model. Includes three original practice questions.
Preview the topics
- A model must learn to prefer short, polite answers over long ones, and for each prompt the team has a preferred answer and a rejected one. Which method fits?
- During a fine-tuning job, the training loss keeps falling, but the validation loss has risen for two epochs. What is happening, and what should the team try?
- A team must compare two fine-tuned checkpoints for a few hours, as cheaply as possible, before choosing one for production. Which deployment type fits?