CoolFace
Modelpublic

shirasko/qwen3.5-2b-crisp-cannabis

sourceHugging Faceupdated 20d agoView on Hugging Face
0likes16downloads
Model Card

Unlearned Checkpoint

FieldValue
Unlearning methodCRISP
Base modelQwen/Qwen3.5-2B
Target conceptCannabis
Checkpoint typeLoRA Adapter
Rank / seed100 / 42
Train eval protocolmc

Unlearning Configuration

Selected hyperparameters (from unlearned_checkpoints.json):

ParameterValue
alpha50
delta_embed0
k_features5
k_features_embed0
layer_hi21
layer_lo5
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)**
Efficacy0.3270.063
Specificity0.7520.927
Harmonic mean0.4550.119
Relearning QA (MC)—0.82

Full Evaluation (baseline → unlearned)

From evaluation/score_comparison.csv:

MetricBaseline (train)After unlearn (train)Baseline (test)**After unlearn (test)**
QA accuracy0.740.580.880.84
QA fraction10.67310.937
SimDom accuracy0.840.70.640.68
SimDom fraction10.76311
MMLU accuracy0.560.480.5880.542
MMLU fraction10.74210.864

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