open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_NPO_lr5e-05_beta0.1_alpha1_epoch10
This model is an instruction-tuned language model from the 'open-unlearning' project, specifically a Llama-3.2-1B variant. It has undergone unlearning using the NPO method with a learning rate of 5e-05, beta of 0.1, alpha of 1, and trained for 10 epochs to forget 10 specific items. This specialized training makes it suitable for research into machine unlearning and evaluating the effectiveness of unlearning techniques on large language models.
Loading preview...
Model Overview
This model, unlearn_tofu_Llama-3.2-1B-Instruct_forget10_NPO_lr5e-05_beta0.1_alpha1_epoch10, is an instruction-tuned variant of the Llama-3.2-1B architecture. It is part of the 'open-unlearning' initiative, focusing on the application and evaluation of machine unlearning techniques.
Key Characteristics
- Base Model: Llama-3.2-1B-Instruct.
- Unlearning Method: NPO (Neural Process Optimization) was applied.
- Unlearning Parameters: Configured with a learning rate of 5e-05, beta of 0.1, alpha of 1, and trained for 10 epochs.
- Forgetting Target: Designed to forget 10 specific data points or concepts.
Intended Use Cases
This model is primarily intended for research and development in the field of machine unlearning. It can be used to:
- Evaluate the efficacy of the NPO unlearning method.
- Compare unlearning performance against other techniques.
- Study the impact of different unlearning hyperparameters.
- Develop and test new metrics for assessing successful unlearning in LLMs.