Reproduce “Manufacturing Differential Sparsity: Planner-Hardware Co-Design for Efficient World Model Planning” 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

Wenbo Zhu, Paul Yi-Chia Chen, Yuanhan Li, Zishen Wan, Christopher Torng

Candelta co-designs a planner and accelerator for repeated world-model planning. It couples candidate action sequences around shared references, letting a term-serial accelerator skip zero bits in activation differences. This preserves each candidate’s sampling marginal but does not reduce predictor calls. Cycle-level simulation reports 1.28–1.80× speedup and 38–52% lower energy than equal-area dense INT8 hardware; Push-T success was comparable.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.