Reproduce “History Forcing: Distilling History-Dependent Continuations for Long-Horizon Video World 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
Hangyi Kuang, Chengzhi Wang, Yefei He, Weijie Wang, Yuke Zhao, Zhuokun Chen, Zisu Huang, Yifan Yang and 3 more
History Forcing transfers history-dependent information to video students for long rollouts and camera-controlled world modeling. It compares an autoregressive teacher’s predictions with and without extra history, then adds their difference to a separate distillation target. On 60-second VBench-Long videos, its 1.3B student scores 79.82 versus 79.22 for Rolling Forcing; this reported comparison does not isolate the correction mechanism.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.