The open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_NPO_lr5e-05_beta0.1_alpha2_epoch5 model is a 1 billion parameter instruction-tuned language model based on the Llama-3.2 architecture. This model is specifically designed for machine unlearning, focusing on forgetting specific information or behaviors. Its primary use case involves research and applications in data privacy, model editing, and mitigating undesirable outputs by selectively removing learned knowledge.
Loading preview...
Overview
This model, unlearn_tofu_Llama-3.2-1B-Instruct_forget10_NPO_lr5e-05_beta0.1_alpha2_epoch5, is a 1 billion parameter instruction-tuned language model built upon the Llama-3.2 architecture. It is specifically developed for machine unlearning, a process aimed at selectively removing specific information or behaviors from a trained model without retraining from scratch. The model's name indicates its focus on forgetting 10 items using the NPO (Negative Preference Optimization) method with specific learning rate, beta, alpha, and epoch settings.
Key Capabilities
- Targeted Forgetting: Designed to unlearn specific data points or patterns.
- Instruction Following: Retains instruction-following capabilities post-unlearning.
- Research in Unlearning: Serves as a valuable tool for exploring and advancing machine unlearning techniques.
Good for
- Data Privacy Research: Investigating methods to remove sensitive data from models.
- Model Editing: Modifying model behavior or knowledge without full retraining.
- Mitigating Bias/Harmful Content: Developing strategies to remove undesirable learned information.
- Understanding Model Memory: Studying how models retain and forget information.