About This Task

By setting the pipeline output type, you make sure the Playground displays results in the most relevant way. The platform can auto-detect the type, but this may not always be accurate. To avoid issues, we recommend setting it explicitly in YAML.

If you set a type manually, it overrides the auto-detected one. For example, if your pipeline is generative but you set pipeline_output_type: "chat", the Playground will use the chat view.

Available Types

You can set pipeline_output_type to one of these values:

• generative: For pipelines where an LLM generates new text as a response.
• chat: For conversational pipelines.
• extractive: For pipelines that extract answers directly from documents.
• document: For pipelines that return full documents or multiple documents as results.

How Playground Adapts

Expand each section to check how the Playground interface changes for each pipeline output type:

Generative pipelines

For generative pipelines, the query input field is at the top of the page and the generated answers show below it:

Chat pipelines

Playground adjusts to the chat view:

When the user starts the conversation, it keeps previous questions and answers in view:

Extractive pipelines

For extractive pipelines, the query input field is at the top and the extracted answers appear below it:

Document search pipelines

For document search pipelines, the query input field is at the top:

Once the user asks a query, the related documents are displayed on the page:

Set the Pipeline Output Type

1. Log in to deepset AI Platform and go to Pipelines.

2. Find the pipeline you want to run async, click More Actions next to it, and choose Edit.

   

3. When the pipeline opens in Pipeline Builder, switch to the YAML view.

   

4. Open the Query tab and add the following line at the beginning of your pipeline YAML:

pipeline_output_type: "<PIPELINE_TYPE>"
# Choose one of these options:
# pipeline_output_type: "generative"
# pipeline_output_type: "extractive"
# pipeline_output_type: "chat"
# pipeline_output_type: "document"

5. Save your pipeline.

Example

What To Do Next

You can deploy your pipeline and try it out in Playground.