Add tutorial for federated fine-tuning of language models with Huggin…

[omid-jf]

Summary

Add a tutorial for fine-tuning a HuggingFace text-classification model (IMDb dataset) using Federated Learning with OpenFL.

Type of Change (Mandatory)

Specify the type of change being made.

• Adding a new tutorial

Description (Mandatory)

This PR introduces a tutorial for performing federated learning on a text classification task using HuggingFace library and the experimental workflow API of OpenFL. It uses the IMDb dataset for binary classification.

The tutorial walks through:

• Installing required dependencies
• Implementing the Federated Averaging (fed_avg) function
• Building the federated workflow (FLSpec)
• Running a local simulation using LocalRuntime
• Setting up a production-ready federated infrastructure
• Deploying and running the same workflow via FederatedRuntime

Testing

• Executed the full tutorial on a clean environment (Python 3.11)
• Verified local simulation workflow completes successfully
• Verified FederatedRuntime deployment works as intended
• Achieves the metrics below for simulation and deployment modes:

It is shown in the screenshot above that after 2 rounds of federated learning, the aggregated model achieves similar F1 score (~0.6) to the local models.

[psfoley]

Thanks for the contribution, @omid-jf! We will take some time to review this and get back to you in the next day or two