ClaudioSavelli/FAME_GA_llama32-1b-instruct-qa
ClaudioSavelli/FAME_GA_llama32-1b-instruct-qa is a 1 billion parameter instruction-tuned language model, derived from meta-llama/Llama-3.2-1B-Instruct, with a 32768 token context length. This model has been unlearned using the Gradient Ascent method within the FAME setting. It is specifically designed for research into model unlearning techniques and their effects on instruction-following capabilities.
Loading preview...
Model Overview
ClaudioSavelli/FAME_GA_llama32-1b-instruct-qa is a 1 billion parameter instruction-tuned model, built upon the meta-llama/Llama-3.2-1B-Instruct architecture. It features a substantial context length of 32768 tokens, making it suitable for processing longer inputs.
Key Characteristics
- Unlearning Method: This model has undergone an "unlearning" process using the Gradient Ascent (GA) method, specifically within the FAME (Forgetting by Adversarial Model Editing) setting.
- Base Model: It originates from the
meta-llama/Llama-3.2-1B-Instructseries, indicating a foundation in Llama-3.2's instruction-following capabilities. - Research Focus: The primary purpose of this model is to facilitate research into model unlearning techniques, particularly how Gradient Ascent impacts a model's learned knowledge and instruction adherence.
Use Cases
- Investigating Unlearning: Ideal for researchers studying the effectiveness and implications of model unlearning methods like Gradient Ascent.
- Evaluating Forgetting: Useful for analyzing how specific data or behaviors can be removed from a model's knowledge base.
- Understanding Model Robustness: Can be used to explore the resilience of instruction-tuned models against unlearning interventions.
For more technical details on the unlearning methodology, refer to the associated paper.