JoaoBoer/tofu_Llama-3.1-8B-Instruct_forget05_SatImp

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

JoaoBoer/tofu_Llama-3.1-8B-Instruct_forget05_SatImp is an 8 billion parameter instruction-tuned Llama-3.1 model, developed by JoaoBoer, that has undergone unlearning using the SatImp method on the TOFU 'forget05' split. This model is specifically designed for research in machine unlearning, serving as a baseline or draft model within the Speculative-Decoding-Unlearning project. Its primary characteristic is its demonstrated ability to forget specific information while retaining general utility, as indicated by its unlearning metrics.

Loading preview...

Overview

JoaoBoer/tofu_Llama-3.1-8B-Instruct_forget05_SatImp is an 8 billion parameter instruction-tuned Llama-3.1 model that has been subjected to machine unlearning. Developed by JoaoBoer, this model was unlearned on the TOFU forget05 dataset split using the SatImp method, within the open-unlearning framework. It functions as a weight-unlearning baseline or draft model for the Speculative-Decoding-Unlearning project.

Key Capabilities

  • Targeted Forgetting: Demonstrates specific unlearning of information from the forget05 split of the TOFU dataset.
  • Research Baseline: Serves as a foundational model for experiments in machine unlearning and speculative decoding.
  • Llama-3.1 Architecture: Benefits from the robust capabilities of the Llama-3.1 instruction-tuned base model.

Unlearning Metrics

The model exhibits specific performance on unlearning metrics, including:

  • exact_memorization: 0.5742
  • extraction_strength: 0.0668
  • forget_Q_A_PARA_Prob: 0.0675
  • forget_Q_A_gibberish: 0.8529
  • model_utility: 0.6024

Good For

  • Researchers studying machine unlearning techniques.
  • Developing and testing new unlearning algorithms.
  • Experiments involving speculative decoding with unlearned models.
  • Understanding the trade-offs between forgetting and model utility.

Finetunes
1 models