ClaudioSavelli/FAME_GA_llama32-1b-1p25-instruct-qa

TEXT GENERATIONConcurrency Cost:1Model Size:1BQuant:BF16Ctx Length:32kPublished:Apr 30, 2026License:otherArchitecture:Transformer Cold

ClaudioSavelli/FAME_GA_llama32-1b-1p25-instruct-qa is a 1 billion parameter instruction-tuned model, derived from meta-llama/Llama-3.2-1b-Instruct, with a 32768 token context length. This model has been specifically unlearned using the Gradient Ascent method within the FAME (Forgetting A Model Effectively) setting. Its primary use case is for research and experimentation in model unlearning and privacy-preserving AI, demonstrating how specific information can be removed post-training.

Loading preview...

Overview

ClaudioSavelli/FAME_GA_llama32-1b-1p25-instruct-qa is a 1 billion parameter instruction-tuned model built upon the meta-llama/Llama-3.2-1b-Instruct architecture, featuring a substantial 32768 token context length. Its key differentiator lies in its development using the Gradient Ascent (GA) method for model unlearning within the FAME (Forgetting A Model Effectively) setting. This process aims to remove specific learned information from the model post-training.

Key Capabilities

  • Model Unlearning Research: Specifically designed for experiments and studies related to the FAME setting and Gradient Ascent unlearning techniques.
  • Instruction Following: Retains instruction-following capabilities from its base Llama-3.2-1b-Instruct model.
  • Long Context Processing: Benefits from the 32768 token context length for handling extensive inputs.

Good for

  • Researchers and practitioners exploring model unlearning methodologies.
  • Evaluating the effectiveness of Gradient Ascent in removing specific data or behaviors from LLMs.
  • Developing and testing privacy-preserving AI techniques.
  • Understanding the impact of unlearning on model performance and generalization.

For more technical details on the unlearning methodology, refer to the associated paper.