From the course: AI Toolkit Essentials for Visual Studio Code

Opening playgrounds - Visual Studio Code Tutorial

From the course: AI Toolkit Essentials for Visual Studio Code

Start my 1-month free trial Buy for my team

Opening playgrounds

“

- [Instructor] So far in this course, we have looked at the model catalog and various ways to import models into AI toolkit. However, we have not in any meaningful way actually used one. We're going to start with the Playground. Now the Playground provides us with a simple interface to interact with a model directly. This can give us some basic verification that we have connected to the model correctly and it's working. It can also tell us a little about the performance of the model in the environment we are trying to run it in. This is particularly important if we are running the model locally. Our machine may not be powerful enough to interact with it effectively, and that can have an impact if we plan on running any test of the model. We can also add a deployed instance of a model and make sure that the hardware's running on platform satisfactory. So we're going to start by using Claude model that we added earlier. So we're going to go to the Claude and right click on it and Load in Playground. So this is a model that will allow us to chat with it since it's a multimodal model, so we're also going to be able to send other collateral. The two main features of the window we see are the prompt bar and a View Code button. So let's start with View Code. This window shows us the Python code needed to create the model with any parameters we may have changed. However, changing things in this window does not change any settings with the Playground. It is informational only. However, we can save the file if we want to use it as sample code to run the model. So let's close that back up. Now let's try to work with the prompt in the model. We can do that at the prompt bar at the bottom of the window. And sometimes if Visual Studio Code is not full screen, I find this little prompt bar is hidden and you can't see it or scroll to it. And so if you're in trouble seeing this prompt bar, you may want to make Visual Studio Code full screen. So let's type in something to send off to Claude, so, "Hello Claude." Okay, it's able to respond to us in a meaningful way and let's try following up. And so we'll go, "Do you have another name?" So yes, it's able to respond meaningfully to our prompts where we asked it for another name and it said, "No, I don't have one," so this model appears to be working correctly. Now in multimodal models, we can do things like attach a file and ask about its contents. So I'll try a picture and ask it what was in the picture. Okay, to add a picture, what we're going to do is go to the bottom and click this little picture icon right here. And we'll add this picture in and type the prompt, "What is in this picture?" and press the Send button. All right, seems like I understood that pretty well. And we can see that there's some other things like the little paper clip there, and that allows us to select other types of files. The web icon allows us to tell the model to search the web. So that's what that one's for. Here we can see that Claude doesn't support this type of search. This tells us that the prompt interface, as would be expected, appears mostly to be the same for all models, though there may be some model-specific features like thinking. So just because an option is presented does not mean it'll work with the loaded model. However, there is a way to filter based on the features the model does actually support. So if we click this button right here, we can use Claude 3.7 Sonnet native supported AI features, and it will filter the features in this prompt. And as you can see, the option to go out and use the web is gone. Now we have a pretty good way not only to test the model, but also to see what input features the current model supports.

Contents