AI Paper Atlas

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

Transport and Merge: Cross-Architecture Merging for Large Language Models

Chenhang Cui, Binyun Yang, Fei Shen, Yuxin Chen, Jingnan Zheng, Xiang Wang, An Zhang, Tat-Seng Chua

ICML 2026 · Large language models › Language, safety and interpretability · approach: Multilingual LLM data and culture · 0 citations (OpenAlex)

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
apply OT-guided fusion with adaptationMalaysian-LLaMA-3-1B-Instruct · MalayMMLU (Humanities)accuracy (%): 42.30 -> 48.81Improved
apply OT-guided fusion with adaptationMalaysian-LLaMA-3-1B-Instruct · MalayMMLU (Social Science)accuracy (%): 40.92 -> 46.91Improved
add post-fusion residual-frozen adaptation (ablation)Fused model · MalayMMLU (Humanities)accuracy (%): 46.19 -> 48.81Improved
add post-fusion adaptation (ablation)Fused model · ARC (Indonesian)accuracy (%): 24.0 -> 23.7Hurt
add post-fusion adaptation (ablation)Fused model · Indonesian benchmarksaverage accuracy (%): 39.18 -> 39.30Improved
apply fusion without adaptationLLaMA-3-1B-Indonesian-QLoRA · Indonesian benchmarksaverage accuracy (%): 38.43 -> 39.18Improved
apply fusion with adaptationLLaMA-3-1B-Instruct · CMMLU Cantoneseaverage accuracy (%): 25.26 -> 27.44Improved
apply fusion without adaptationTyphoon2-1B-Instruct · MGSM (Thai)score: 0.50 -> 0.64Improved
apply fusion without adaptationTyphoon2-1B-Instruct · MMLU (Thai)score: 0.15 -> 0.13Hurt
add adaptation after fusion (ablation)Fused model · Thai benchmarksaverage score: 0.45 -> 0.50Improved
fuse general model into finance-specialized target without adaptationLLaMA-3-1B-TEL-A-finance · finance MMLU subsetsaverage score: 0.32 -> 0.34Improved
apply fusion with adaptationmedical LLaMA-3-1B · medical MMLU subsetsaverage accuracy (%): 46.30 -> 47.87Improved

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