AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Don't Throw Away Your Pretrained Model

Shangbin Feng, Wenhao Yu, Yike Wang, Hongming Zhang, Yulia Tsvetkov, Dong Yu

ICLR 2026 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add trained switcher to alternate pretrained, finetuned, aligned models by segmentaligned Tulu-3-8B · TruthfulQAtask score: 29.01 -> 39.22Improved
add trained switcher (switch-task)aligned Tulu-3-8B · GSM8ktask score: 56.80 -> 59.60Improved
add trained switcher (switch-task)aligned Tulu-3-8B · BBHtask score: 35.20 -> 58.30Improved
add trained switcher (switch-task)aligned Tulu-3-8B · Alpacatask score: 57.46 -> 56.89Hurt
add trained switcher (switch-task)aligned Tulu-3-8B · Poemtask score: 77.70 -> 70.25Hurt
use trained switcher instead of greedy soupgreedy soup · BBHtask score: 36.50 -> 58.30Improved
switcher outperforms all baselines and individual modelsswitch generation · 18 datasetstask score: best on 13 of 18 datasets, 12.9% average relative improvementMixed
fine-tune the switcher instead of random switching (ablation)switch-task · TruthfulQAtask score: 27.07 -> 39.22Improved
fine-tune the switcher instead of untuned switching (ablation)switch-task · BBHtask score: 41.80 -> 58.30Improved
random switching instead of trained switcher (ablation)switch-task · BBHtask score: 58.30 vs 53.10Mixed
use patch size 50 instead of 100 (ablation)switch-task · GSM8ktask score: 44.70 -> 59.60Improved
use smaller patch sizes (20) instead of 50 (ablation)switch-task · Pluralismtask score: 53.20 -> 56.80Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy