tofu_Llama-3.2-3B-Instruct_forget01_UNDIAL
open-unlearning/tofu_Llama-3.2-3B-Instruct_full unlearned on the TOFU forget01 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 | 0.4065 |
| extraction_strength | 0.0413 |
| forget_Q_A_PARA_Prob | 0.0742 |
| forget_Q_A_gibberish | 0.9228 |
| forget_quality | 0.2657 |
| forget_truth_ratio | 0.7394 |
| mia_loss | 0.0759 |
| mia_min_k | 0.0838 |
| mia_min_k_plus_plus | 0.1762 |
| mia_zlib | 0.0431 |
| model_utility | 0.5352 |
| privleak | 107.0621 |