Reproduce “PAR: Progress-Aligned Recurrence for Length Generalization in State Tracking” 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
Pan Wang, Yihao Hu, Guangyuan Dong, Yitong Lin, Hang Wang, Zimi Liu, Zhihao Wen, Jianshe Li and 3 more
PAR trains a recurrent language model to track five items through swaps, supervising each iteration with the state after the corresponding moves and anchoring state norms. On swap-S5, Qwen2.5-0.5B trained on at most 32 moves achieves 95.5% mean full-state accuracy at 512 moves, 16 times the training length. The result concerns state tracking, not open-ended reasoning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.