tofu_Llama-3.2-1B-Instruct_forget10_SimNPO
open-unlearning/tofu_Llama-3.2-1B-Instruct_full unlearned on the TOFU forget10 split with SimNPO, 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: 0.125
alpha: 1
retain_loss_type: NLL
delta: 1
beta: 3.5
TOFU summary metrics
| metric | value |
|---|---|
| exact_memorization | 0.5864 |
| extraction_strength | 0.0559 |
| forget_Q_A_PARA_Prob | 0.0518 |
| forget_Q_A_gibberish | 0.9089 |
| forget_quality | 0.2441 |
| forget_truth_ratio | 0.6676 |
| mia_loss | 0.2687 |
| mia_min_k | 0.2541 |
| mia_min_k_plus_plus | 0.2631 |
| mia_zlib | 0.2266 |
| model_utility | 0.4939 |
| privleak | 20.8110 |