Exploring Gretel's Model Playground (Full Tour with GPT & Tabular LLM)

Speaker 1 (00:00):

Hey, welcome to the Gretto Playground. I'm Nicole. I'm excited to show you what you can do with our new model playground. This is the GPT playground. You can see we're using Gretto GPT model for inference. Now new and updated with state of the Art LAMA two LLM. In this input section, this is where I'll ask the model questions or ask it to generate data. So one quick and simple example we can use here on the right is asking it a question. What is synthetic data and why is it useful? There are other examples we have here for you to get started. Now, after I click generate, results usually take just a few seconds to come back. Sometimes it may take a little longer. So thank you for your patience during the beta field. We really appreciate your feedback. So please consider submitting feedback here based on your playground experience.

(00:49)
You can see here these results were pretty quick to come back. It's really rich answer. I really like that. The answer highlights some use cases for synthetic data, including being cost effective, improving privacy and security. I can also ask the model other questions like if I'm creating a synthetic consumer packaged goods dataset, maybe I'll ask something like, what are some common fields in a typical CPG dataset or other questions like that. But I can also ask the GPT model to generate data. So let's say we've heard from a lot of customers that you're interested in generating massive amounts of text data. So in this case, this is a pretty generic generate an email subject line celebrating a company milestone. Okay, I see the output. It's not quite what I'm interested in. Maybe I'm actually interested in something. Let me change the prompt here. Generate annual subject line, celebrating a launch of a new AI feature. So something like that maybe.

(02:06)
Okay, I can keep on clicking generate if I am interested in seeing what else the model could give me. Maybe this is pretty good, but I'm not sure if this is exactly what I want. If I do like this output and I want more records to generate a much larger dataset, we do have an option to generate batch data based on the prompt. So this flow, you select a project just like you do with other gretto models. Enter the number of records you want. Let's say I want a hundred records, and based on that, if I click generate, this will start a job that will give me a data set of a hundred records based on this prompt.

(02:46)
There's a lot that you can do with a Gretto GPT playground, including of course asking the model questions as well as generating batch data like this. Next, I'll show you what you can do with the gretto tabular LM playground. Okay, the tabular LLM playground. So tabular LM is an early access preview only. You can see here we're in the creating structured data mode and we can enter a text prompt or SQL prompt. So for a text prompt, let's say I want to create a sample employee data set. Now this is a pretty short prompt. I can always add more to the prompt by indicating specific cities or specific aspects of the data. But you can see really quickly the model starts giving us results back and very consistent with the columns that were here in the prompt. I'll show you another SQL query here. Maybe I'm interested in orders database.

(03:43)
So looking at fulfillment data, if I am an online shop or I'm in the e-commerce industry. So again, results are coming back in really quickly and I see again, everything is in line with the headers that I asked for. Up here in the SQL query, we want you to be able to see that you can play around with any queries really quickly. Any prompts see the results come back quick. And if they're not to your liking, you can stop the generation, change the prompt and try again. So this playground is meant to be a very interactive interface for you to see what kind of data is possible from tabular LLM based on the prompt. If I am happy with this, I can also request a batch data workflow. So the way that that would work is that once I'm checking out these results and I'm happy with it, I can say, okay, now instead of this preview of results, I want to generate more data.

(04:43)
So very similar to other gradual workflows, I can create or select an existing project. So I'll create a new project about orders dataset, I can edit how many rows I want, and then this prompt will be used to generate more results. So as a quick look at the Tableau LLM playground, we're really excited for you to try this experience, whether in SQL or in text. And please do make sure to submit feedback. Your feedback is so important to us, especially in this beta phase, and we are excited to see what you do with the playground. Thanks.

‍