JoaoBoer/tofu_Llama-3.2-1B-Instruct_forget10_NPO

TEXT GENERATIONPricing:Input $0.108 / Output $0.804Concurrent Unit Cost:1Model Size:1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 10, 2026License:llama3.2Architecture:Transformer Featherless Exclusive Cold

The JoaoBoer/tofu_Llama-3.2-1B-Instruct_forget10_NPO is a 1 billion parameter Llama-3.2-Instruct model, unlearned on the TOFU forget10 split using the NPO method. Developed within the open-unlearning framework, this model serves as a weight-unlearning baseline for the Speculative-Decoding-Unlearning project. It is specifically designed to explore and evaluate unlearning techniques, demonstrating controlled forgetting of specific data while retaining general utility.

Loading preview...

Model Overview

This model, tofu_Llama-3.2-1B-Instruct_forget10_NPO, is a 1 billion parameter Llama-3.2-Instruct variant that has undergone an unlearning process. It was trained using the open-unlearning framework, specifically applying the NPO (Negative Preference Optimization) method to unlearn the forget10 split of the TOFU dataset. This makes it a key component and baseline model for the Speculative-Decoding-Unlearning research project.

Key Characteristics

  • Unlearning Focus: The primary differentiator is its application of unlearning techniques, aiming to remove specific information (the forget10 split) from the model's knowledge base.
  • NPO Method: Utilizes the NPO method for weight-unlearning, with specific hyperparameters (gamma: 1.0, alpha: 2, retain_loss_type: NLL, beta: 0.1) detailed in its configuration.
  • Evaluation Metrics: Comprehensive TOFU evaluation metrics are provided, including exact_memorization (0.5675), extraction_strength (0.0620), forget_quality (0.8134), and model_utility (0.5463), which quantify the effectiveness of the unlearning process and the model's retained utility.

Use Cases

This model is particularly suited for:

  • Research in Machine Unlearning: Ideal for researchers studying methods to remove data from trained models, evaluating the trade-offs between forgetting and utility.
  • Baseline for Speculative Decoding Unlearning: Serves as a foundational model for projects exploring speculative decoding in the context of unlearning.
  • Understanding Model Forgetting: Provides a practical example for analyzing how specific unlearning techniques impact model behavior and knowledge retention.

Finetunes
1 models