Run Train Create a file, eg, my_trainer.py and use examples and modify your configs to run the file.

Train a Text Inversion using Unlearn Canvas dataset

Before finetuning the model you need to generate safetensors.

from mu.algorithms.forget_me_not.algorithm import ForgetMeNotAlgorithm
from mu.algorithms.forget_me_not.configs import (
    forget_me_not_train_ti_mu,
)

algorithm = ForgetMeNotAlgorithm(
    forget_me_not_train_ti_mu,
    ckpt_path="models/diffuser/style50",
    raw_dataset_dir=(
        "data/quick-canvas-dataset/sample"
    ), 
    steps=10,
    template_name = "Abstractionism", #concept to erase
    dataset_type = "unlearncanvas" ,
    use_sample = True, #train on sample dataset
    output_dir = "outputs/forget_me_not/finetuned_models" #output dir to save finetuned models
)
algorithm.run(train_type="train_ti")

Running the Script in Offline Mode

WANDB_MODE=offline python my_trainer_ti.py

1. Perform Unlearning using Unlearn Canvas dataset

Before running the train_attn script, update the ti_weights_path parameter in the configuration file to point to the output generated from the Text Inversion (train_ti.py) stage

from mu.algorithms.forget_me_not.algorithm import ForgetMeNotAlgorithm
from mu.algorithms.forget_me_not.configs import (
    forget_me_not_train_attn_mu,
)

algorithm = ForgetMeNotAlgorithm(
    forget_me_not_train_attn_mu,
    ckpt_path="models/diffuser/style50",
    raw_dataset_dir=(
        "data/quick-canvas-dataset/sample"
    ),
    steps=10,
    ti_weights_path="outputs/forget_me_not/finetuned_models/Abstractionism/step_inv_10.safetensors",
    template_name = "Abstractionism", #concept to erase
    dataset_type = "unlearncanvas" ,
    use_sample = True, #train on sample dataset
    output_dir = "outputs/forget_me_not/finetuned_models" #output dir to save finetuned models
)
algorithm.run(train_type="train_attn")

Running the Script in Offline Mode

WANDB_MODE=offline python my_trainer_attn.py

How It Works * Default Values: The script first loads default values from the train config file as in configs section.

• Parameter Overrides: Any parameters passed directly to the algorithm, overrides these configs.

• Final Configuration: The script merges the configs and convert them into dictionary to proceed with the training.

This method involves two stages:

1. Train a Text Inversion: The first stage involves training a Text Inversion. Refer to the script train_ti.py for details and implementation. It uses train_ti_config.yaml as config file.

2. Perform Unlearning: The second stage uses the outputs from the first stage to perform unlearning. Refer to the script train_attn.py for details and implementation. It uses train_attn_config.yaml as config file.