Reproduce “UniSpark: Exploring Hybrid Attention for Multimodal Understanding and Generation” 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

Jiancheng Pan, Yuzheng Liu, Hengyuan Cao, Qiyuan Zhang, Yangyan Li, Biao Gong, Zheng Zhang, Ka Long Cheng and 2 more

UniSpark presents a unified model for image generation, understanding, and editing. It combines full attention with GatedDeltaNet linear attention, routing text and semantic visual features separately from continuous visual latents. Noisy image predictions remain private to later blocks, while completed images enter shared context. It scores 2470.3 on MME-S and 0.83 on GenEval; these benchmark results do not isolate the mechanism.

Summary by alphaXiv.

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