Views
No views yet
| Field | Value |
|---|---|
| Unlearning method | PISCES |
| Base model | google/gemma-2-2b-it |
| Target concept | Gambling |
| Checkpoint type | Full Model Weights |
| Rank / seed | 100 / 42 |
| Train eval protocol | open |
unlearned_checkpoints.json):| Parameter | Value |
|---|---|
delta_embed | 0 |
k_features_embed | 0 |
k_pisces | 0.3 |
n_tokens_edited | 0 |
ratio_thresh | 2 |
value_pisces | 36 |
| Metric | Train (after unlearning) | Test (after unlearning) |
|---|---|---|
| Efficacy | 0.889 | 0.75 |
| Specificity | 0.962 | 0.865 |
| Harmonic mean | 0.924 | 0.803 |
evaluation/score_comparison.csv:| Metric | Baseline (train) | After unlearn (train) | Baseline (test) | After unlearn (test) |
|---|---|---|---|---|
| QA accuracy | 0.54 | 0.06 | 0.64 | 0.16 |
| QA fraction | 1 | 0.111 | 1 | 0.25 |
| SimDom accuracy | 0.64 | 0.64 | 0.76 | 0.62 |
| SimDom fraction | 1 | 1 | 1 | 0.816 |
| MMLU accuracy | 0.52 | 0.5 | 0.551 | 0.527 |
| MMLU fraction | 1 | 0.926 | 1 | 0.92 |
| File | Description |
|---|---|
unlearned_checkpoints.json | Checkpoint metadata & hyperparameters |
evaluation/evaluation_summary.json | Full evaluation payload (train/test/relearning) |
evaluation/score_comparison.csv | Baseline vs. unlearned comparison table |