Unlearn Diff

Unlearn Diff is an open-source Python package designed to streamline the development of unlearning algorithms and establish a standardized evaluation pipeline for diffusion models. It provides researchers and practitioners with tools to implement, evaluate, and extend unlearning algorithms effectively.

Documentation

You can find the full documentation for this project at the url given below. https://ramailotech.github.io/msu_unlearningalgorithm/

Features

• Comprehensive Algorithm Support: Includes commonly used concept erasing and machine unlearning algorithms tailored for diffusion models. Each algorithm is encapsulated and standardized in terms of input-output formats.

• Automated Evaluation: Supports automatic evaluation on datasets like UnlearnCanvas or IP2P. Performs standard and adversarial evaluations, outputting metrics as detailed in UnlearnCanvas and UnlearnDiffAtk.

• Extensibility: Designed for easy integration of new unlearning algorithms, attack methods, defense mechanisms, and datasets with minimal modifications.

Supported Algorithms

The initial version includes established methods benchmarked in UnlearnCanvas and defensive unlearning techniques:

• CA (Concept Ablation)
• ED (Erase Diff)
• ESD (Efficient Substitution Distillation)
• FMN (Forget Me Not)
• SU (Saliency Unlearning)
• SH (ScissorHands)
• SA (Selective Amnesia)
• SPM (Semi Permeable Membrane)
• UCE (Unified Concept Editing) For detailed information on each algorithm, please refer to the respective README.md files located inside mu/algorithms.

Project Architecture

The project is organized to facilitate scalability and maintainability.

• data/: Stores data-related files.

  • i2p-dataset/: contains i2p-dataset

    • sample/: Sample dataset
    • full/: Full dataset

  • quick-canvas-dataset/: contains quick canvas dataset

    • sample/: Sample dataset
    • full/: Full dataset

• docs/: Documentation, including API references and user guides.

• outputs/: Outputs of the trained algorithms.

• examples/: Sample code and notebooks demonstrating usage.

• logs/: Log files for debugging and auditing.

• models/: Repository of trained models and checkpoints.

• mu/: Core source code.

  • algorithms/: Implementation of various algorithms. Each algorithm has its own subdirectory containing code and a README.md with detailed documentation.
    • esd/: ESD algorithm components.
      • README.md: Documentation specific to the ESD algorithm.
      • algorithm.py: Core implementation of ESD.
      • configs/: Configuration files for training and generation tasks.
      • constants/const.py: Constant values used across the ESD algorithm.
      • environment.yaml: Environment setup for ESD.
      • model.py: Model architectures specific to ESD.
      • sampler.py: Sampling methods used during training or inference.
      • scripts/train.py: Training script for ESD.
      • trainer.py: Training routines and optimization strategies.
      • utils.py: Utility functions and helpers.
    • ca/: Components for the CA algorithm.
      • README.md: Documentation specific to the CA algorithm.
      • ...and so on for other algorithms
  • core/: Foundational classes and utilities.
    • base_algorithm.py: Abstract base class for algorithm implementations.
    • base_data_handler.py: Base class for data handling.
    • base_model.py: Base class for model definitions.
    • base_sampler.py: Base class for sampling methods.
    • base_trainer.py: Base class for training routines.
  • datasets/: Dataset management and utilities.
    • __init__.py: Initializes the dataset package.
    • dataset.py: Dataset classes and methods.
    • helpers/: Helper functions for data processing.
    • unlearning_canvas_dataset.py: Specific dataset class for unlearning tasks.
  • helpers/: Utility functions and helpers.
    • helper.py: General-purpose helper functions.
    • logger.py: Logging utilities to standardize logging practices.
    • path_setup.py: Path configurations and environment setup.

• tests/: Test suites for ensuring code reliability.

• stable_diffusion/: Components for stable diffusion.

• lora_diffusion/: Components for the LoRA Diffusion.

Datasets

We use the Quick Canvas benchmark dataset, available here. Currently, the algorithms are trained using 5 images belonging to the themes of Abstractionism and Architectures.

Usage

This section contains the usage guide for the package.

Installation

pip install unlearn_diff

Prerequisities

Ensure conda is installed on your system. You can install Miniconda or Anaconda:

• Miniconda (recommended): https://docs.conda.io/en/latest/miniconda.html
• Anaconda: https://www.anaconda.com/products/distribution

After installing conda, ensure it is available in your PATH by running. You may require to restart the terminal session:

conda --version

Create environment:

create_env <algorithm_name>

eg: create_env erase_diff

Activate environment:

conda activate <environment_name>

eg: conda activate mu_erase_diff

The <algorithm_name> has to be one of the folders in the mu/algorithms folder.

Downloading data and models.

After you install the package, you can use the following commands to download.

1. Dataset:

• i2p:
  • Sample:

  download_data sample i2p
  

  • Full:

  download_data full i2p
  

• quick_canvas:
  • Sample:

  download_data sample quick_canvas
  

  • Full:

  download_data full quick_canvas
  

2. Model:

• compvis:

  download_model compvis
  

• diffuser:

  download_model diffuser
  

Run Train

Each algorithm has their own script to run the algorithm, Some also have different process all together. Follow usage section in readme for the algorithm you want to run with the help of the github repository. You will need to create a train_config.yaml anywhere in your machine, and pass it's path as --config_path parameter.

Here is an example for Erase_diff algorithm.

WANDB_MODE=offline python -m mu.algorithms.erase_diff.scripts.train \
--config_path <path_to_config_in_your_machine>

The default algorithm specific train_config.yaml makes use of the model_config.yaml with default settings. You can also create your own model_config.yaml and update it's path in the train_config.yaml file to tweak the original model parameters. The details about each parameter in config files are written in the readme for each of the algorithm.

NOTE Make sure to update these parameters in train_config.yaml. Otherwise, the train script will not run properly. Also, update other parameters as per your usage.

model_config_path: "configs/erase_diff/model_config.yaml"  # path to model_config.yaml. 
ckpt_path: "models/compvis/style50/compvis.ckpt"  # Checkpoint path for compvis or diffuser model
raw_dataset_dir: "data/i2p-dataset/sample" # path where your dataset was downloaded
processed_dataset_dir: "mu/algorithms/erase_diff/data"  # path to directory, where you want the trained model data to be stored