Reproduce “Composable Interventions for Language Models” 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
Arinbjorn Kolbeinsson, Kyle O'Brien, Tianjin Huang, Shanghua Gao, Shiwei Liu, Jonathan Richard Schwarz, Anurag Vaidya, Faisal Mahmood and 3 more
The authors introduce a framework for testing whether editing, unlearning, and compression preserve one another’s effects and whether application order matters. It defines Order-free Error and Order Sensitivity and evaluates ten methods on Llama-3 8B. In tested settings, compression often made editing and unlearning harder, while scores could conceal order sensitivity; results are limited to the tested models, methods, and tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.