CoolFace
Modelpublic

shirasko/gemma-2-2b-it-snmf-baseball

sourceHugging Faceupdated 2mo agoView on Hugging Face
0likes18downloads
Model Card

Unlearned Checkpoint

FieldValue
Unlearning methodSNMF
Base modelgoogle/gemma-2-2b-it
Target conceptBaseball
Checkpoint typeFull Model Weights
Rank / seed100 / 42
Train eval protocolmc

Unlearning Configuration

Selected hyperparameters (from unlearned_checkpoints.json):

ParameterValue
coverage_thresh0.95
delta_embed0
delta_in1
delta_out10
feature_sourceall
k_features_embed0
k_features_mlp_in33
k_features_mlp_out29
layer_hi_in12
layer_hi_out8
layer_lo_in0
layer_lo_out0
n_tokens_edited0
ratio_thresh2
w_modeboth

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

Headline scores used for checkpoint selection:

MetricTrain (after unlearning)**Test (after unlearning)**
Efficacy0.7460.615
Specificity0.7710.502
Harmonic mean0.7580.553
Relearning QA (MC)—0.46

Full Evaluation (baseline → unlearned)

From evaluation/score_comparison.csv:

MetricBaseline (train)After unlearn (train)Baseline (test)**After unlearn (test)**
QA accuracy0.840.40.640.4
QA fraction10.25410.385
SimDom accuracy0.680.520.740.42
SimDom fraction10.62810.347
MMLU accuracy0.520.540.5510.524
MMLU fraction1110.91

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