课程: Learn Databricks GenAI

Use AI Playground

- [Instructor] So over in the Databricks test environment here, you can see the playground and we have the ability to take a prompt, and I'm going to paste one in here. Act as an expert in Databricks. Generate a table with five key features of the ML playground for a data science audience. And then run it against the default model, which is meta LAMA 3.3, 70 billion instruct. So it's really designed as a test bed for prompt engineering when you're first getting started designing your application. Now in addition to the single endpoint, and here it's returning the results, you can also associate tools and tools basically are functions. You can also export the results. Notice there are suggested questions. There is the ability to add system information and then here's where you configure the temperature, top PE, top K and then there's a new capability called AI judge that allows you to get assessments from the LLM judges and the Mosaic AI agent evaluation. So I'm going to go ahead and clear this out and we're going to not only use this AI judge, but we're going to add an additional endpoint. When we add another endpoint, we can then select a different foundational model, and I'm just going to select the DBRX instruct. Notice there's no ability to add functions here. You're just using a query. You can see that here, there's no tools. And then if you have this sync button highlighted and I paste in this prompt, I can go ahead and run the same prompt in both of these foundational models. So this is a really great way to get started with prompt engineering. Very handy. And in case you're wondering if you can work with additional models like OpenAI, you can, and I'll be talking about how to do that in subsequent movies as an external endpoint. Notice with the AI judge, it gives us information about the content, whether it's safe and relevant and then also the suggested questions are available here. So this is a fantastic way when you are starting to work with custom generative AI to understand which foundational models and tools and what model settings are going to produce the best results as you're building.

内容