watsonx Custom Model for Langflow

This repository contains a custom Langflow component for integrating IBM watsonx.ai foundation models into Langflow. It provides a configurable interface for interacting with watsonx models using Langflow's component system.

Features

• Supports multiple watsonx foundation models.
• Allows fine-grained control over generation parameters like max_tokens, temperature, top_k, and top_p.
• Securely manages API keys with SecretStrInput.
• Provides dropdowns, sliders, and text inputs for easy configuration.

Installation

Ensure you have Python 3.8+ installed, then install the required dependencies:

pip install langflow langchain_ibm ibm-watsonx-ai pydantic

Usage

Setting Up the Component in Langflow

1. Clone this repository:

   git clone <repository-url>
   cd <repository>

2. Create the Custom Component Directory Structure:

   /app/custom_components/
   └── llms/
       └── watsonx_component.py
   

3. Configure Langflow to Recognize the Custom Component:

   • Set the LANGFLOW_COMPONENTS_PATH environment variable to point to your custom components directory:

     export LANGFLOW_COMPONENTS_PATH=/app/custom_components/

   • Alternatively, set this variable within your Python script:

     import os
     os.environ['LANGFLOW_COMPONENTS_PATH'] = '/app/custom_components/'

4. Launch Langflow:

   langflow run --components-path=/app/custom_components/ 

   • Access the Langflow interface at http://localhost:7860.

5. Utilize the watsonx Component in Your Workflow:

   • Locate the "LLMs" category in the components sidebar.
   • Drag and drop the "watsonx" component into your workflow canvas.
   • Configure the component’s parameters, such as model_name, api_key, and url.
   • Connect the watsonx component to other components in your workflow.

6. Execute and Test Your Workflow:

   • Run the workflow to ensure the watsonx component functions correctly.
   • Monitor the outputs and adjust configurations as needed.

Configuration

Parameter	Description
model_name	Selectable dropdown of supported watsonx models.
api_key	Secret API key for authentication.
url	Endpoint URL for watsonx API.
project_id	Project ID associated with the API key.
max_tokens	Maximum number of tokens to generate.
temperature	Controls randomness; higher values increase diversity.
top_k	Number of top tokens to sample from.
top_p	Probability mass cutoff for token selection.

Example Code

Here's an example of how to initialize and use the component:

from custom_component import WatsonxComponent

watsonx_model = WatsonxComponent()
model = watsonx_model.build_model()
response = model.invoke("Hello, watsonx!")
print(response)

Contributing

Feel free to fork this repository and submit pull requests. Make sure to follow best practices and test your changes before submitting.

License

This project is licensed under the MIT License.