Quick start: Prompt a foundation model using Prompt Lab

You need a project to store Prompt Lab assets.

Watch a video to see how to create a sandbox project and associate a service. Then follow the steps to verify that you have an existing project or create a sandbox project.

1. From the watsonx home screen, scroll to the Projects section. If you see any projects listed, then skip to Task 2. If you don't see any projects, then follow these steps to create a project.

2. Click Create a sandbox project. When the project is created, you will see the sandbox project in the Projects section.

For more information or to watch a video, see Creating a project.

To preview this task, watch the video beginning at 00:03.

You can type your prompt text in a freeform, plain text editor and then click Generate to send your prompt to the model. Follow these steps to use the Prompt Lab in Freeform mode:

1. From the home screen, click the Open in Prompt Lab tile.

2. Select each checkbox to accept the acknowledgements, and then click Skip tour.

3. Click the Freeform tab to prompt a foundation model in Freeform mode.

4. Click Switch mode.

5. Copy and paste the following text in the text field, and then click Generate to see the output for the Class name: Problem.

Classify this customer message into one of two classes: question, problem.
Class name: Question
Description: The customer is asking a technical question or a how-to question about our products or services.
Class name: Problem
Description: The customer is describing a problem they are having. They might say they are trying something, but it's not working. They might say they are getting an error or unexpected results.
Message: I'm having trouble registering for a new account.
Class name:

To preview this task, watch the video beginning at 00:19.

You can type your prompt in a structured format. The structured format is helpful for few-shot prompting, when your prompt has multiple examples. Follow these steps to use the Prompt Lab in Structured mode:

1. Click the Structured tab.

2. Click Switch mode.

3. In the Instruction field, copy and paste the following text: Given a message submitted to a customer-support chatbot for a cloud software company, classify the customer's message as either a question or a problem description so the chat can be routed to the correct support team.

4. In the Setup field, copy and paste the following text in each column:

   Values for Setup field

   Input	Output
   When I try to log in, I get an error.	Problem
   Where can I find the plan prices?	Question
   What is the difference between trial and paygo?	Question
   The registration page crashed, and now I can't create a new account.	Problem
   What regions are supported?	Question
   I can't remember my password.	Problem

   
   

5. In the Try field, copy and paste the following text: I'm having trouble registering for a new account.

6. Click Generate to see the output Problem.

To preview this task, watch the video beginning at 00:33. If you’re not sure how to begin, sample prompts can get your started. Follow these steps to use the sample prompts:

1. Open the Sample prompts icon to display the list.
2. Scroll through the list, and click the Marketing email generation sample prompt.
3. View the selected model. When you load a sample prompt, an appropriate model is selected for you.
4. Open the Model Parameters panel . The appropriate decoding and stopping criteria parameters are set automatically too.
5. Click Generate to submit the sample prompt to the model, and see the sample email output.

To preview this task, watch the video beginning at 01:04. You can submit the same prompt to a different model. If you need to generate output in a specific, structured output, such as JSON, HTML, or XML, you can do that by choosing a model that has been fine-tuned for generating structured output, using few-shot examples in your prompt, or tuning a model. Follow these steps to choose a different foundation model:

1. Click Model > View all foundation models.
2. Click a model to learn more about a model, and see detail such as the model architecture, pretraining data, fine-tuning information, and performance against benchmarks.
3. Click Back to return to the list of models.
4. Select either the flan-t5-xxl-11b or mt0-xxl-13b foundation model, and click Select model.
5. Hover over the model output column and click the X icon to delete the previous output.
6. Click the same sample prompt, Marketing email generation, from the list.
7. Click Generate to generate output using the new model.

To preview this task, watch the video beginning at 01:28.

You can experiment with changing decoding or stopping criteria parameters. Follow these steps to adjust model parameters.

The following table defines the model parameters available for the flan-t5-xxl-11b foundation model.

1. Change the Top K parameter to 10 to make it less likely the model will go off topic.
2. Click X to delete the previous model output.
3. Click the same sample prompt from the list.
4. Click Generate to generate output using the new model parameters.
5. Click the Session history icon after submitting multiple prompts to view your session history.
6. Click any entry to work with a previous prompt, model specification, and parameter settings, and then click Restore.
7. Edit the prompt, change the model, or adjust decoding and stopping criteria parameters.
8. Click Generate to generate output using the updated information.

To preview this task, watch the video beginning at 02:15.

You can save your work in three formats:

Follow these steps to save your work:

1. Click Save work > Save as.
2. Select Prompt template.
3. For the name, type Sample prompts.
4. Select the View in project after saving option.
5. Click Save.
6. On the project's Assets tab, click the Sample prompts asset to load that prompt in the Prompt Lab and get right back to work.
7. Click the Saved prompts to see saved prompt from your sandbox project.