open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_GradDiff_lr2e-05_alpha5_epoch10
This is a 1 billion parameter instruction-tuned Llama-3.2 model, developed by open-unlearning, with a context length of 32768 tokens. It is specifically designed for unlearning tasks, having been fine-tuned to forget specific information using the GradDiff method. This model is intended for research and applications requiring selective knowledge removal from large language models.
Loading preview...
Overview
This model is a 1 billion parameter instruction-tuned Llama-3.2 variant, developed by open-unlearning, with a substantial context length of 32768 tokens. It has been fine-tuned using the GradDiff method to perform unlearning, specifically targeting the forgetting of certain information. This makes it distinct from standard LLMs by offering capabilities for selective knowledge removal.
Key Capabilities
- Instruction-tuned Llama-3.2 base: Benefits from the foundational capabilities of the Llama-3.2 architecture.
- Unlearning via GradDiff: Engineered to forget specific data points or concepts, making it suitable for privacy-preserving or bias mitigation research.
- Large context window: Supports processing and generating long sequences of text, up to 32768 tokens.
Good for
- Research in machine unlearning: Ideal for exploring methods and effects of removing information from LLMs.
- Privacy-preserving AI: Can be used in scenarios where specific sensitive data needs to be expunged from a trained model.
- Mitigating model bias: Potentially useful for reducing biases by unlearning problematic associations or data.