Pipeline Builder now supports enforced output types for LLM nodes

[krandoing]

Pipeline Builder now supports enforced output types for LLM nodes

You can now specify the output types of your Use LLM nodes in Pipeline Builder! These functionalities ensure consistent output types, enhancing your ability to maintain data type consistency and optimize data processing within your pipelines.

How to enforce output type in the Use LLM node:

1. Ensure that your Foundry frontend instance is updated to v6.321.0 or higher.

2. Add a Use LLM node to your pipeline. If you use a template, submit the template first to get to the prompt page.

3. On the prompt screen, locate the Output type field and select the desired output type from the dropdown menu.

Use the dropdown menu to specify an output type for your Use LLM node.

For entity extraction, you can also specify the output types inside a struct to provide more flexibility and control over your data outputs.

Example of a struct output type containing an array, a string, and an integer.

Learn more about using LLMs in Pipeline Builder from our documentation.

[smitshahngra]

How do we use the output type field for LLM block in Typescript code?

const systemMessage = 'Test message'
const gptResponse = await GPT_4o.createChatCompletion({
            messages: [systemMessage],
            params: { temperature: 0.0 },
});

Is there any parameter or API exposed in Typescript packages? Or is there an ETA for the same?

[helenq]

Currently, the enforced output type feature in the LLM block is not exposed or usable as an API outside of Pipeline Builder