Reproduce “Don't Start Over: A Cost-Effective Framework for Migrating Personalized Prompts Between LLMs” 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
Ziyi Zhao, Chongming Gao, Yang Zhang, Haoyan Liu, Weinan Gan, Huifeng Guo, Yong Liu, Fuli Feng
PUMA migrates personalized recommendation prompts when a shared language model is replaced. It learns a shared adapter that maps each user’s frozen soft prompt into the target model’s embedding space, using a selected subset of users for training. On Amazon Movies & TV, PUMA achieved 0.9135 RMSE versus 0.9414 for full retraining and reduced reported training time 50×, but evidence is limited to recommendation migration.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.