JoaoBoer/tofu_Llama-3.1-8B-Instruct_forget10_IdkNLL
The JoaoBoer/tofu_Llama-3.1-8B-Instruct_forget10_IdkNLL is an 8 billion parameter instruction-tuned language model, based on the Llama-3.1 architecture, specifically unlearned on the TOFU 'forget10' split using the IdkNLL method. Developed by JoaoBoer, this model serves as a weight-unlearning baseline within the Speculative-Decoding-Unlearning project. Its primary characteristic is its focus on demonstrating and evaluating machine unlearning techniques, particularly for privacy and data retention control.
Loading preview...
Model Overview
The tofu_Llama-3.1-8B-Instruct_forget10_IdkNLL is an 8 billion parameter instruction-tuned model derived from the Llama-3.1 architecture. It has undergone a specific unlearning process on the TOFU forget10 dataset split, utilizing the IdkNLL method. This model is developed by JoaoBoer as a baseline for weight-unlearning within the Speculative-Decoding-Unlearning research project.
Key Capabilities & Characteristics
- Machine Unlearning Demonstration: This model is specifically designed to showcase and evaluate machine unlearning techniques, focusing on the removal of specific data (the 'forget10' split of the TOFU dataset).
- IdkNLL Method: It implements the IdkNLL (I don't know NLL) method for unlearning, a technique aimed at selectively forgetting information from the model's weights.
- Research Baseline: Serves as a foundational model for further research into speculative decoding and unlearning methodologies.
- Llama-3.1 Architecture: Built upon the robust Llama-3.1 instruction-tuned base, providing a strong foundation for language understanding and generation, even after unlearning.
Good for
- Research in Machine Unlearning: Ideal for researchers and developers exploring privacy-preserving AI, data deletion, and model editing techniques.
- Evaluating Unlearning Methods: Provides a concrete example and baseline for assessing the effectiveness of different unlearning algorithms.
- Understanding Model Forgetting: Useful for studying how models forget specific information and the impact on overall model utility and privacy metrics.