tofu_Llama-3.2-3B-Instruct_forget05_UNDIAL
open-unlearning/tofu_Llama-3.2-3B-Instruct_full unlearned on the TOFU forget05 split with UNDIAL, 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: 1
beta: 10
retain_loss_type: NLL
TOFU summary metrics
| metric | value |
|---|---|
| exact_memorization | 0.3480 |
| extraction_strength | 0.0349 |
| forget_Q_A_PARA_Prob | 0.1099 |
| forget_Q_A_gibberish | 0.8206 |
| forget_quality | 0.0002 |
| forget_truth_ratio | 0.5943 |
| mia_loss | 0.2997 |
| mia_min_k | 0.5575 |
| mia_min_k_plus_plus | 0.5769 |
| mia_zlib | 0.2253 |
| model_utility | 0.5606 |
| privleak | -30.8137 |