Modelpublic

shirasko/qwen3.5-2b-rmu-golf

sourceHugging Faceupdated 23d agoView on Hugging Face

0likes369downloads

Model Card

Unlearned Checkpoint

Selected hyperparameters (from unlearned_checkpoints.json):

Headline scores used for checkpoint selection:

From evaluation/score_comparison.csv:

Metric	Baseline (train)	After unlearn (train)	Baseline (test)	After unlearn (test)
QA accuracy	0.82	0.48	0.74	0.44
QA fraction	1	0.404	1	0.388
SimDom accuracy	0.72	0.54	0.74	0.52
SimDom fraction	1	0.617	1	0.551
MMLU accuracy	0.56	0.48	0.588	0.565
MMLU fraction	1	0.742	1	0.932

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