Databricks · Databricks Generative AI Engineer Associate
Choosing Models
Databricks Generative AI Engineer Associate · Lesson 7 of 17 · 49:20
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 Databricks Certified Generative AI Engineer Associate exam. In this lesson: the attributes of a chat model (quality, context window, latency, cost, tool use, languages), embedding models and their context windows, the same embedding model for documents and queries, model cards and licenses, and choosing by measurement with MLflow runs. Includes three original practice questions.
Preview the topics
- Copperline splits its manuals into chunks of about 2,000 tokens, and embeds them with a model whose window is 512 tokens. Answers often miss steps that appear late in a section. What is the most likely cause?
- Copperline found an open model that looks good in a demo. Before building on it, the team wants to know its intended use, its known limitations, and whether the license allows commercial use. Where should the team look first?
- Copperline evaluated three chat models on the same evaluation set, and logged each run in MLflow. The two larger models score the same as the medium one on quality, but cost more and answer more slowly. Which model should the team keep?