CoolFace
Modelpublic

shirasko/llama-3.1-8b-instruct-rmu-gambling

sourceHugging Faceupdated 2mo agoView on Hugging Face
0likes16downloads
Model Card

Unlearned Checkpoint

FieldValue
Unlearning methodRMU
Base modelmeta-llama/Llama-3.1-8B-Instruct
Target conceptGambling
Checkpoint typeFull Model Weights
Rank / seed200 / 42
Train eval protocolmc

Unlearning Configuration

Selected hyperparameters (from unlearned_checkpoints.json):

ParameterValue
alpha50
delta_embed0
k_features_embed0
layer_id7
layer_ids5,6,7
lr0.0003
n_tokens_edited0
param_ids6
setting_nameS1lid7L567
steering30

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

Headline scores used for checkpoint selection:

MetricTrain (after unlearning)**Test (after unlearning)**
Efficacy11
Specificity0.9460.909
Harmonic mean0.9720.952
Relearning QA (MC)—0.78

Full Evaluation (baseline → unlearned)

From evaluation/score_comparison.csv:

MetricBaseline (train)After unlearn (train)Baseline (test)**After unlearn (test)**
QA accuracy0.840.020.920.02
QA fraction1010
SimDom accuracy0.980.9410.92
SimDom fraction10.94510.893
MMLU accuracy0.620.60.650.62
MMLU fraction10.94610.925

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