JoaoBoer/tofu_Llama-3.1-8B-Instruct_forget01_IdkNLL
The JoaoBoer/tofu_Llama-3.1-8B-Instruct_forget01_IdkNLL is an 8 billion parameter instruction-tuned Llama 3.1 model, developed by JoaoBoer, that has undergone unlearning on the TOFU 'forget01' split using the IdkNLL method. This model is designed as a weight-unlearning baseline within the Speculative-Decoding-Unlearning project, focusing on evaluating unlearning effectiveness. It features a 32768-token context length and is primarily used for research into machine unlearning and privacy-preserving AI.
Loading preview...
Model Overview
The tofu_Llama-3.1-8B-Instruct_forget01_IdkNLL is an 8 billion parameter instruction-tuned Llama 3.1 model, developed by JoaoBoer, specifically modified for machine unlearning research. It was created by applying an unlearning process to the open-unlearning/tofu_Llama-3.1-8B-Instruct_full model on the TOFU forget01 dataset split, utilizing the IdkNLL method within the open-unlearning framework.
This model serves as a weight-unlearning baseline and a draft model for the Speculative-Decoding-Unlearning project, aiming to explore and evaluate the efficacy of unlearning techniques. It maintains a substantial context length of 32768 tokens.
Key Capabilities
- Targeted Unlearning: Demonstrates unlearning on specific data (TOFU
forget01split) using the IdkNLL method. - Research Baseline: Provides a foundational model for further experimentation in machine unlearning and privacy.
- Llama 3.1 Architecture: Benefits from the robust capabilities of the Llama 3.1 instruction-tuned base model.
Good for
- Machine Unlearning Research: Ideal for researchers studying methods to remove specific information from trained models.
- Privacy-Preserving AI: Useful for exploring techniques to enhance data privacy in large language models.
- Comparative Studies: Can be used to compare different unlearning algorithms and their impact on model utility and forgetting metrics.