JoaoBoer/tofu_Llama-3.1-8B-Instruct_forget05_SimNPO

TEXT GENERATIONPricing:Input $0.2 / Cached $0.028 / Output $0.32Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 10, 2026License:llama3.1Architecture:Transformer Featherless Exclusive Cold

The JoaoBoer/tofu_Llama-3.1-8B-Instruct_forget05_SimNPO is an 8 billion parameter Llama-3.1-Instruct model, unlearned on the TOFU 'forget05' split using the SimNPO method. Developed within the open-unlearning framework, it serves as a weight-unlearning baseline for the Speculative-Decoding-Unlearning project. This model is specifically designed to demonstrate and evaluate machine unlearning capabilities, particularly in forgetting specific data while retaining general utility.

Loading preview...

Model Overview

This model, tofu_Llama-3.1-8B-Instruct_forget05_SimNPO, is an 8 billion parameter variant of the Llama-3.1-Instruct architecture. It has undergone a specific unlearning process using the SimNPO method on the TOFU forget05 dataset split. This work is part of the open-unlearning framework and is utilized as a weight-unlearning baseline within the Speculative-Decoding-Unlearning project.

Key Characteristics

  • Unlearning Focus: Specifically trained to forget information from the TOFU forget05 split.
  • Methodology: Employs the SimNPO (Simplified Negative Preference Optimization) unlearning technique.
  • Baseline Model: Serves as a foundational model for research into speculative decoding and unlearning.
  • Evaluation Metrics: Detailed TOFU summary metrics are provided, including exact_memorization (0.5672), forget_quality (0.5453), and model_utility (0.5085), indicating its performance in both forgetting and retaining general capabilities.

Use Cases

This model is primarily intended for:

  • Research in Machine Unlearning: Ideal for studying the effectiveness and mechanisms of unlearning algorithms.
  • Evaluation of Unlearning Techniques: Provides a benchmark for comparing different unlearning methods.
  • Development of Privacy-Preserving AI: Useful for exploring how models can selectively forget sensitive or outdated information.

Finetunes
1 models