Keshigeyan Chandrasegaran, Yue Zhao, Armin W. Thomas, Ehsan Adeli, Jiajun Wu, Juan Carlos Niebles, Li Fei-Fei, Michael Poli · 2026 · arXiv (Cornell University)
DOI: 10.48550/arxiv.2610.26863 · cited by 0 works (OpenAlex) · open-access PDF
Sistem teks-ke-gambar (T2I) modern menggunakan pra-latihan bahasa skala internet untuk pembuatan gambar berbasis teks. Kami bertanya apakah pra-latihan bahasa diperlukan untuk generasi T2I yang kuat. Kami memperkenalkan Halcyon, sebuah keluarga model autoregresif multimodal awal-fusion yang pretrained sepenuhnya dari awal pada GPIC, sebuah dataset 100M gambar-tekst pasangan. Halcyon-0.52B melampaui LlamaGen-XL/8 (48.0 vs. 63.3 gFD) dengan 18.2$\times$ kurang pra-latihan komputasi. Selanjutnya, Halcyon-1.6B melampaui JiT-T2I-XXL/8++ (35.3 vs. 41.0F gD) dengan 275$\$times kurang pra-latihan komputasi. Eksperimen terkontrol menunjukkan bahwa pra-latihan teks tidak
Machine translation (Meta M2M100); the English original follows.
Modern text-to-image (T2I) systems use internet-scale language pretraining for text-conditioned image generation. We ask whether language pretraining is necessary for strong T2I generation. We introduce Halcyon, a family of early-fusion multimodal autoregressive models pretrained entirely from scratch on GPIC, a dataset of 100M image-text pairs. Halcyon-0.52B outperforms LlamaGen-XL/8 (48.0 vs. 63.3 gFD) with 18.2$\times$ less pretraining compute. Further, Halcyon-1.6B outperforms JiT-T2I-XXL/8++ (35.3 vs. 41.0 gFD) with 275$\times$ less pretraining compute. Controlled experiments show that text-only pretraining does not improve T2I generation. Halcyon also learns image captioning by replacing caption dropout with I2T supervision, maintaining competitive T2I generation quality under the same pretraining budget. We show that strong T2I systems can be trained from scratch without internet-scale language pretraining. Code and models are available at https://cs.stanford.edu/~keshik/halcyon
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
Topics: Closed captioning, Computer science, Artificial intelligence, Scratch, Natural language processing, Language model
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.26863, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange