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

Unlearned Checkpoint

Field	Value
Unlearning method	CRISP
Base model	Qwen/Qwen3.5-2B
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	10
delta_embed	0
k_features	10
k_features_embed	0
layer_hi	15
layer_lo	5
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	0.314	0.392
Specificity	0.856	0.872
Harmonic mean	0.459	0.541
Relearning QA (MC)	—	0.6

---

Full Evaluation (baseline → unlearned)

From evaluation/score_comparison.csv:

Metric	Baseline (train)	After unlearn (train)	Baseline (test)	After unlearn (test)
QA accuracy	0.76	0.6	0.76	0.56
QA fraction	1	0.686	1	0.608
SimDom accuracy	0.7	0.66	0.76	0.68
SimDom fraction	1	0.911	1	0.843
MMLU accuracy	0.56	0.5	0.588	0.555
MMLU fraction	1	0.806	1	0.902

---

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