open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_GradDiff_lr1e-05_alpha5_epoch10

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:1BQuant:BF16Ctx Length:32kPublished:May 15, 2025Architecture:Transformer Warm

The open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_GradDiff_lr1e-05_alpha5_epoch10 model is a 1 billion parameter instruction-tuned language model based on the Llama-3.2 architecture. This model has undergone an unlearning process to remove specific information, making it suitable for applications requiring controlled knowledge retention or removal. With a context length of 32768 tokens, it is designed for tasks where selective forgetting of data is crucial.

Loading preview...

Model Overview

This model, open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_GradDiff_lr1e-05_alpha5_epoch10, is a 1 billion parameter instruction-tuned language model built upon the Llama-3.2 architecture. It features a substantial context length of 32768 tokens, enabling it to process and generate longer sequences of text.

Key Characteristics

  • Unlearning Process: This model has been specifically subjected to an "unlearning" procedure, indicated by unlearn_tofu and forget10_GradDiff. This suggests it has been trained to remove or "forget" certain data points or knowledge, likely using a Gradient Difference (GradDiff) method.
  • Instruction-Tuned: The Instruct in its name signifies that it has been fine-tuned to follow instructions effectively, making it suitable for conversational AI, question answering, and command execution.
  • Llama-3.2 Base: Built on the Llama-3.2 family, it inherits the foundational capabilities and architectural strengths of this series.

Potential Use Cases

  • Controlled Information Access: Ideal for scenarios where a model needs to operate without specific sensitive or outdated information.
  • Privacy-Preserving AI: Could be used in applications requiring the removal of personally identifiable information (PII) or proprietary data from a model's knowledge base.
  • Research in Unlearning: Serves as a valuable tool for researchers exploring methods and effects of machine unlearning in large language models.