shirasko/gemma-2-2b-it-crisp-uranium

Unlearned Checkpoint

Field	Value
Unlearning method	CRISP
Base model	google/gemma-2-2b-it
Target concept	Uranium
Checkpoint type	LoRA Adapter
Rank / seed	100 / 42
Train eval protocol	mc

---

Unlearning Configuration

Selected hyperparameters (from unlearned_checkpoints.json):

Parameter	Value
alpha	50
delta_embed	0
k_features	10
k_features_embed	0
layer_hi	20
layer_lo	4
layer_step	2
lora_rank	4
lr	0.0001
n_tokens_edited	0
num_epochs	2

---

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

Headline scores used for checkpoint selection:

Metric	Train (after unlearning)	Test (after unlearning)
Efficacy	0.769	1
Specificity	0.719	0.72
Harmonic mean	0.743	0.837
Relearning QA (MC)	—	0.4

---

Full Evaluation (baseline → unlearned)

From evaluation/score_comparison.csv:

Metric	Baseline (train)	After unlearn (train)	Baseline (test)	After unlearn (test)
QA accuracy	0.64	0.34	0.6	0.24
QA fraction	1	0.231	1	0
SimDom accuracy	0.62	0.48	0.74	0.54
SimDom fraction	1	0.622	1	0.592
MMLU accuracy	0.52	0.48	0.551	0.527
MMLU fraction	1	0.852	1	0.92

---

Files in This Repository

File	Description
unlearned_checkpoints.json	Checkpoint metadata & hyperparameters
evaluation/evaluation_summary.json	Full evaluation payload (train/test/relearning)
evaluation/score_comparison.csv	Baseline vs. unlearned comparison table