Reproduce “Text-to-Image Generative Modeling Without Language Pretraining” 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
Keshigeyan Chandrasegaran, Yue Zhao, Armin W. Thomas, Ehsan Adeli, Jiajun Wu, Juan Carlos Niebles, Li Fei-Fei, Michael Poli
Halcyon trains a text-to-image model from scratch on paired image-caption examples, jointly modeling caption and image tokens without a web-scale text-pretrained generator. On GPIC at 256×256 after one epoch, Halcyon-1.6B reaches 35.28 gFD versus 40.97 for JiT-T2I-XXL/8++, using 275× less total pretraining compute; controlled tests find no benefit from text-only pretraining in evaluated settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.