CoolFace
Modelpublic

shirasko/llama-3.1-8b-instruct-pisces-uranium

sourceHugging Faceupdated 2mo agoView on Hugging Face
0likes13downloads
Model Card

Unlearned Checkpoint

FieldValue
Unlearning methodPISCES
Base modelmeta-llama/Llama-3.1-8B-Instruct
Target conceptUranium
Checkpoint typeFull Model Weights
Rank / seed200 / 42
Train eval protocolopen

Unlearning Configuration

Selected hyperparameters (from unlearned_checkpoints.json):

ParameterValue
delta_embed0
k_features_embed0
k_pisces0.2
n_tokens_edited0
ratio_thresh2
value_pisces10

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

Headline scores used for checkpoint selection:

MetricTrain (after unlearning)**Test (after unlearning)**
Efficacy0.7880.906
Specificity0.7020.792
Harmonic mean0.7420.845

Full Evaluation (baseline → unlearned)

From evaluation/score_comparison.csv:

MetricBaseline (train)After unlearn (train)Baseline (test)**After unlearn (test)**
QA accuracy0.60.140.640.06
QA fraction10.21210.094
SimDom accuracy0.720.40.70.46
SimDom fraction10.54110.657
MMLU accuracy0.620.680.650.649
MMLU fraction1110.998

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