open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_NPO_lr5e-05_beta0.1_alpha2_epoch10

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

The open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_NPO_lr5e-05_beta0.1_alpha2_epoch10 model is a 1 billion parameter instruction-tuned language model, likely based on the Llama-3.2 architecture. This model is specifically designed for "unlearning" certain information, indicated by "unlearn_tofu" and "forget10" in its name, suggesting it has undergone a process to remove specific data points or knowledge. Its primary differentiator lies in its unlearning capabilities, making it suitable for use cases requiring selective knowledge removal or privacy-preserving AI.

Loading preview...

Model Overview

This model, open-unlearning/unlearn_tofu_Llama-3.2-1B-Instruct_forget10_NPO_lr5e-05_beta0.1_alpha2_epoch10, is an instruction-tuned language model with approximately 1 billion parameters. While specific details on its architecture and training are marked as "More Information Needed" in the provided model card, its naming convention strongly suggests a focus on unlearning or forgetting specific information. The "unlearn_tofu" and "forget10" components indicate that the model has been subjected to a process designed to remove or reduce its knowledge of certain data, likely related to the "TOFU" dataset, and aiming to forget 10 specific items or concepts.

Key Characteristics

  • Unlearning Focus: The primary characteristic is its capability for selective knowledge removal, making it distinct from standard instruction-tuned models.
  • Instruction-Tuned: It is designed to follow instructions, typical of many modern LLMs.
  • Parameter Count: With 1 billion parameters, it falls into the smaller, more efficient category of language models.

Potential Use Cases

  • Privacy-Preserving AI: Ideal for scenarios where specific sensitive information needs to be removed from a model's knowledge base post-training.
  • Data Governance: Useful for complying with data retention policies or "right to be forgotten" requests by selectively erasing learned data.
  • Research in Machine Unlearning: Serves as a valuable tool for studying and experimenting with different machine unlearning techniques and their effectiveness.