Playground LLM

Estimated reading: 4 minutes

Playground LLM is a Language Model component in Robility Flow that allows users to select and switch between supported Language Models directly from the Playground.

It retrieves the available models based on the providers configured for the Project in Robility Flow. The selected model is initialized and provided to the connected Agent for task execution. This allows different Language Models to be used with the same Agent configuration without modifying the existing workflow.

Prerequisites

Before using Playground LLM, complete the following:

a. Add the required LLM providers and configure their models on the Model Providers page in Robility Flow.
b. Add the Playground LLM component to the flow canvas.
c. Connect the Language Model output of Playground LLM to the Language Model input of an Agent.

Note: The model selector and Refresh option appear in the Playground after the Playground LLM component is added to the flow canvas and connected to an Agent.

How It Fits into Robility Flow

Playground LLM acts as the Language Model source for an Agent. It provides the selected Language Model through its output, which the connected Agent uses to interpret instructions and perform the assigned task.

Configuration

Playground LLM includes two optional input parameters for specifying a fallback model.

Inputs

Input Description
Fallback Model Name Specifies the model to use when no Language Model is selected in the Playground.
Fallback Provider Specifies the provider associated with the fallback model, such as OpenAI, Anthropic, Groq, or vLLM.

Note: Both inputs are optional. When configured, they provide an alternative model configuration if no Playground model is selected.

Output

Output Description
Language Model Returns the initialized Language Model. Connect this output to the Language Model input of an Agent.

How It Works

Playground LLM uses the model selected in the Playground to initialize the Language Model for the connected Agent.

The process involves the following steps:

1. Load Available Models: Retrieves the models configured under the available providers within the applicable Project.
2. Select a Model: Displays the available models in the Playground selector.
3. Retrieve Model Configuration: Retrieves the provider configuration and credentials associated with the selected model.
4. Initialize the Model: Initializes the selected Language Model using the retrieved configuration.
5. Provide the Output: Passes the initialized Language Model to the connected Agent for execution.

Fallback Behavior

If no model is selected in the Playground, the component uses the configured fallback model and provider, if available. If neither a Playground model nor a fallback configuration is available, the component raises an error.

Model Refresh

The Refresh option reloads the available models in the Playground.

Use this option after updating model configurations in Robility Flow to retrieve the latest available models.

Note: Refresh is not available when the flow is accessed through a Shareable Playground.

Execution Scenarios

Playground LLM supports two execution scenarios: Normal Playground Execution and Shareable Playground Execution.

Normal Playground Execution

When a flow containing Playground LLM is opened in the Playground, the model selector displays the models configured under the providers available in the applicable Project in Robility Flow.

Select a model to use it for the current execution. The Refresh option can be used to reload the available models when required.

Shareable Playground Execution

When a shared Playground is accessed, Playground LLM identifies the original flow associated with the shared URL.

It retrieves the original flow owner’s details and uses the associated Project and model configurations to initialize the selected Language Model. This ensures that model selection uses the configuration associated with the original flow.

Limitation: The Refresh option is not available when accessing a flow through a Shareable Playground.

Example Use Case

Switching Language Models for Document Summarization

Consider a workflow that uses an Agent to summarize uploaded documents. Different Language Models can be selected to perform the summarization task.

The model can be changed between executions to compare how different Language Models handle the same summarization task.

Configuration

1. Add the required providers and model configurations on the Model Providers page.
2. Add Playground LLM to the flow canvas.
3. Connect its Language Model output to the Language Model input of an Agent.

Execution

1. Open the workflow in the Playground.
2. Upload a document for summarization.
3. Select an available model from the model selector.
4. Run the workflow.
5. Select another available model and run the workflow again, if required.

For each execution, Playground LLM initializes the selected model and provides it to the Agent. The Agent processes the uploaded document using the selected Language Model.

Error Handling

The following table describes the possible scenarios during model selection and initialization.

Scenario Behavior
Invalid Playground model configuration Attempts to use the configured fallback model, if available.
Checkpoint configuration unavailable Attempts to retrieve the model configuration from the memory cache.
Flow owner unavailable Raises an error if the original flow owner cannot be identified.

Note: Successful model initialization requires the appropriate provider configuration and credentials.

Share this Doc

Playground LLM

Or copy link

CONTENTS
Robility Chatbot
Robility Assistant
Online