JoaoBoer/tofu_Llama-3.1-8B-Instruct_forget05_IdkNLL
The JoaoBoer/tofu_Llama-3.1-8B-Instruct_forget05_IdkNLL is an 8 billion parameter instruction-tuned language model based on Llama-3.1, specifically unlearned on the TOFU 'forget05' split using the IdkNLL method. Developed by JoaoBoer within the open-unlearning framework, this model serves as a weight-unlearning baseline for the Speculative-Decoding-Unlearning project. It demonstrates specific unlearning characteristics, with an exact memorization score of 0.8042 and a forget_Q_A_PARA_Prob of 0.0850, making it suitable for research into model unlearning and privacy-preserving AI.
Loading preview...
Overview
The tofu_Llama-3.1-8B-Instruct_forget05_IdkNLL is an 8 billion parameter instruction-tuned model derived from Llama-3.1. It has undergone a specific unlearning process on the TOFU forget05 dataset split using the IdkNLL method, developed within the open-unlearning framework. This model functions as a weight-unlearning baseline and a draft model for the Speculative-Decoding-Unlearning project.
Key Capabilities
- Targeted Unlearning: Specifically designed to forget information from the TOFU
forget05split, demonstrating anexact_memorizationof 0.8042 and aforget_Q_A_PARA_Probof 0.0850. - Research Baseline: Serves as a foundational model for exploring and evaluating unlearning techniques, particularly within the context of speculative decoding.
- Privacy-Preserving AI: Relevant for research into mitigating privacy risks by selectively removing learned information from large language models.
Good for
- Machine Unlearning Research: Ideal for researchers studying methods to remove specific data from trained models.
- Privacy-Aware LLM Development: Useful for experiments in creating models with reduced memorization of sensitive or unwanted information.
- Comparative Analysis: Can be used to compare the effectiveness of different unlearning algorithms and their impact on model utility and forgetting metrics.