tofu_Llama-3.2-1B-Instruct_forget10_NPO
open-unlearning/tofu_Llama-3.2-1B-Instruct_full unlearned on the TOFU forget10 split with NPO, trained with the open-unlearning framework. Used as a weight-unlearning baseline / draft model in the Speculative-Decoding-Unlearning project.
Full training config: .hydra/config.yaml. TOFU evaluation outputs: evals/.
Method hyperparameters
gamma: 1.0
alpha: 2
retain_loss_type: NLL
beta: 0.1
TOFU summary metrics
| metric | value |
|---|---|
| exact_memorization | 0.5675 |
| extraction_strength | 0.0620 |
| forget_Q_A_PARA_Prob | 0.0362 |
| forget_Q_A_gibberish | 0.9299 |
| forget_quality | 0.8134 |
| forget_truth_ratio | 0.6412 |
| mia_loss | 0.1898 |
| mia_min_k | 0.1838 |
| mia_min_k_plus_plus | 0.2168 |
| mia_zlib | 0.2013 |
| model_utility | 0.5463 |
| privleak | 32.2032 |