tofu_Llama-3.2-1B-Instruct_forget01_NPO
open-unlearning/tofu_Llama-3.2-1B-Instruct_full unlearned on the TOFU forget01 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 | 0.6310 |
| extraction_strength | 0.0651 |
| forget_Q_A_PARA_Prob | 0.0733 |
| forget_Q_A_gibberish | 0.9303 |
| forget_quality | 0.5786 |
| forget_truth_ratio | 0.6597 |
| mia_loss | 0.2956 |
| mia_min_k | 0.2456 |
| mia_min_k_plus_plus | 0.3075 |
| mia_zlib | 0.3294 |
| model_utility | 0.5467 |
| privleak | 42.5030 |