gemma-2-2b-it-snmf-wmdp-bio – AI Model by shirasko | AlphaNeural AI

You can deploy this model and start earning money today!

shirasko/
gemma-2-2b-it-snmf-wmdp-bio

like

0

text-generation

google/gemma-2-2b-it

text-generation-inference

endpoints_compatible

Views

No views yet

Unlearned Checkpoint

Field	Value
Unlearning method	SNMF
Base model	`google/gemma-2-2b-it`
Target concept	wmdp-bio
Checkpoint type	Full Model Weights
Rank / seed	100 / 42
Train eval protocol	mc

Unlearning Configuration

Selected hyperparameters (from unlearned_checkpoints.json):

Parameter	Value
`coverage_thresh`	0.95
`delta_embed`	0
`delta_in`	4
`delta_out`	4
`feature_source`	all
`k_features_embed`	0
`k_features_mlp_in`	141
`k_features_mlp_out`	17
`layer_hi_in`	25
`layer_hi_out`	17
`layer_lo_in`	0
`layer_lo_out`	9
`n_tokens_edited`	0
`ratio_thresh`	2
`w_mode`	both

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

Headline scores used for checkpoint selection:

Metric	Train (after unlearning)	Test (after unlearning)
Efficacy	0.816	0.634
Specificity	0.912	0.72
Harmonic mean	0.861	0.674
Relearning QA (MC)	—	0.48

Full Evaluation (baseline → unlearned)

From evaluation/score_comparison.csv:

Metric	Baseline (train)	After unlearn (train)	Baseline (test)	After unlearn (test)
QA accuracy	0.74	0.34	0.66	0.4
QA fraction	1	0.184	1	0.366
SimDom accuracy	0.62	0.56	0.76	0.58
SimDom fraction	1	0.838	1	0.647
MMLU accuracy	0.52	0.56	0.551	0.494
MMLU fraction	1	1	1	0.811

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