JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget01_UNDIAL

TEXT GENERATIONPricing:Input $0.2036 / Output $1.34Concurrent Unit Cost:1Model Size:3.2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 10, 2026License:llama3.2Architecture:Transformer Featherless Exclusive Cold

The JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget01_UNDIAL is a 3.2 billion parameter instruction-tuned Llama-3.2 model, specifically a version of `open-unlearning/tofu_Llama-3.2-3B-Instruct_full` that has undergone unlearning on the TOFU `forget01` split using the UNDIAL method. Developed within the open-unlearning framework, this model serves as a weight-unlearning baseline for the Speculative-Decoding-Unlearning project. It is optimized for evaluating and demonstrating machine unlearning capabilities, particularly in reducing memorization of specific data while retaining general utility.

Loading preview...

Model Overview

This model, tofu_Llama-3.2-3B-Instruct_forget01_UNDIAL, is a 3.2 billion parameter instruction-tuned variant of the Llama-3.2 architecture. It is derived from open-unlearning/tofu_Llama-3.2-3B-Instruct_full and has been specifically processed using the UNDIAL method to unlearn information from the TOFU forget01 dataset split. This unlearning process was conducted within the open-unlearning framework.

Key Characteristics & Purpose

  • Machine Unlearning Baseline: It functions as a weight-unlearning baseline or draft model for the Speculative-Decoding-Unlearning project.
  • Unlearning Metrics: Evaluation metrics highlight its unlearning performance, with exact_memorization at 0.4065 and forget_Q_A_gibberish at 0.9228, indicating a significant reduction in memorization of the forgotten data while maintaining a model_utility of 0.5352.
  • Methodology: The unlearning process utilized specific hyperparameters including gamma: 1.0, alpha: 1, beta: 10, and retain_loss_type: NLL.

Ideal Use Cases

  • Research in Machine Unlearning: Excellent for researchers studying and developing machine unlearning techniques.
  • Evaluating Unlearning Algorithms: Can be used to benchmark and compare different unlearning methods against a known baseline.
  • Privacy-Preserving AI: Relevant for applications requiring models to forget specific sensitive data while retaining general knowledge.

Finetunes
1 models