Reproduce “Dual-Slot: From Serial Loops to Parallel Slots in 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

Yiqin Hao, Zitong Wang, Boyi Zeng, He Li, Shixiang Song, Feichen Song, Yifan Liu, Ran Tao and 2 more

Dual-Slot reuses shared layers while shifting loop work across adjacent tokens: the current prediction uses the previous token’s latent state as the current token’s latent state is computed. In 1B dense models, Pile validation perplexity is 7.00 versus 7.25 for Vanilla, and decoding throughput retains 93.1–96.5% of Vanilla’s; A0.6B MoE retains 98.1–99.6% in tested settings.

Summary by alphaXiv.

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