CoolFace
Modelpublic

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

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

Unlearned Checkpoint

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

Unlearning Configuration

Selected hyperparameters (from unlearned_checkpoints.json):

ParameterValue
alpha100
delta_embed0
k_features_embed0
layer_id11
layer_ids9,10,11
lr0.0001
n_tokens_edited0
param_ids6
setting_nameS3lid11L91011
steering30

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

Headline scores used for checkpoint selection:

MetricTrain (after unlearning)**Test (after unlearning)**
Efficacy0.0920.196
Specificity0.9180.959
Harmonic mean0.1680.326
Relearning QA (MC)—0.84

Full Evaluation (baseline → unlearned)

From evaluation/score_comparison.csv:

MetricBaseline (train)After unlearn (train)Baseline (test)**After unlearn (test)**
QA accuracy0.90.840.760.66
QA fraction10.90810.804
SimDom accuracy0.780.70.780.76
SimDom fraction10.84910.962
MMLU accuracy0.620.640.650.632
MMLU fraction1110.955

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