Microsoft · AI-300
Tuning, pipelines, and distributed training
AI-300 · Lesson 5 of 15 · 31:36
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: sweep jobs with Choice and continuous distributions, grid, random, and Bayesian sampling, Bandit, median stopping, and truncation selection policies, pipelines and schedules, and distributed PyTorch training. Includes three original practice questions.
Preview the topics
- A sweep job finishes, but no trial shows a value for its primary metric, accuracy. The training script logs a metric named val_accuracy. What is the fix?
- In a pipeline, the training step fails, right after a two-hour data preparation step succeeded. Where should the team look first?
- A PyTorch job must train on two nodes with four GPUs each, with one process per GPU. Which settings fit?