Reproduce “LiST: Local-Simplex Test-Time LoRA Fusion” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yihua Shao, Jia Li, Siyu Chen, Xinyu Luo, Yang Liu, Kecheng Chen, Xinwei Long, Lingyu Zhu and 7 more
LiST searches input-specific blends of frozen LoRA adapters, using a target-conditioned neighborhood and a label-free energy score with feasibility checks; it falls back to a prior blend when the search is unreliable. On Qwen3-VL-8B, its average across four unseen-task benchmark scores is 68.18, versus 64.31 for MEMO, using seen-domain adapters without a target-domain adapter.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.