CoolFace
Modelpublic

shirasko/gemma-2-2b-it-crisp-uranium

sourceHugging Faceupdated 2mo agoView on Hugging Face
0likes7downloads
Model Card

Unlearned Checkpoint

FieldValue
Unlearning methodCRISP
Base modelgoogle/gemma-2-2b-it
Target conceptUranium
Checkpoint typeLoRA Adapter
Rank / seed100 / 42
Train eval protocolmc

Unlearning Configuration

Selected hyperparameters (from unlearned_checkpoints.json):

ParameterValue
alpha50
delta_embed0
k_features10
k_features_embed0
layer_hi20
layer_lo4
layer_step2
lora_rank4
lr0.0001
n_tokens_edited0
num_epochs2

Primary Unlearning Metrics (held-out test, MC protocol)

Headline scores used for checkpoint selection:

MetricTrain (after unlearning)**Test (after unlearning)**
Efficacy0.7691
Specificity0.7190.72
Harmonic mean0.7430.837
Relearning QA (MC)—0.4

Full Evaluation (baseline → unlearned)

From evaluation/score_comparison.csv:

MetricBaseline (train)After unlearn (train)Baseline (test)**After unlearn (test)**
QA accuracy0.640.340.60.24
QA fraction10.23110
SimDom accuracy0.620.480.740.54
SimDom fraction10.62210.592
MMLU accuracy0.520.480.5510.527
MMLU fraction10.85210.92

Files in This Repository

FileDescription
unlearned_checkpoints.jsonCheckpoint metadata & hyperparameters
evaluation/evaluation_summary.jsonFull evaluation payload (train/test/relearning)
evaluation/score_comparison.csvBaseline vs. unlearned comparison table