open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_NPO_lr5e-05_beta0.5_alpha5_epoch10
The open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_NPO_lr5e-05_beta0.5_alpha5_epoch10 model is a 1 billion parameter instruction-tuned language model based on the Llama-3.2 architecture. This model is specifically designed for unlearning, focusing on forgetting specific information (forget10) using the NPO method with particular hyperparameters. Its primary differentiator lies in its application of unlearning techniques to a Llama-based instruction model, making it suitable for research into model privacy and data removal.
Loading preview...
Model Overview
This model, unlearn_tofu_Llama-3.2-1B-Instruct_forget10_NPO_lr5e-05_beta0.5_alpha5_epoch10, is a 1 billion parameter instruction-tuned variant of the Llama-3.2 architecture. It has been developed with a specific focus on "unlearning" capabilities, aiming to remove or forget certain information from its training data. The model's name indicates its configuration: it targets forgetting 10 specific data points (forget10) using the Negative Preference Optimization (NPO) method, with a learning rate of 5e-05, beta of 0.5, alpha of 5, and trained for 10 epochs.
Key Characteristics
- Architecture: Based on the Llama-3.2 family, providing a strong foundation for instruction following.
- Parameter Count: A compact 1 billion parameters, making it efficient for deployment and research.
- Unlearning Focus: Explicitly trained with unlearning objectives, specifically targeting the removal of 10 data points.
- Methodology: Utilizes the NPO (Negative Preference Optimization) technique for the unlearning process.
- Context Length: Supports a context length of 32768 tokens, allowing for processing longer inputs.
Potential Use Cases
- Research in Machine Unlearning: Ideal for academic and industrial research into methods for removing specific data from trained models.
- Privacy-Preserving AI: Can serve as a baseline or component for developing models that can adapt to data removal requests.
- Model Auditing: Useful for understanding how unlearning techniques impact model behavior and performance.
Limitations
As indicated by the model card, many details regarding its development, training data, evaluation, and specific biases are currently marked as "More Information Needed." Users should exercise caution and conduct thorough evaluations before deploying this model in sensitive applications, especially concerning the completeness and effectiveness of its unlearning capabilities.