HuggingFace Daily Papers·· 2 天前
Halcyon：不依赖语言预训练的从零训练文生图模型
Text-to-Image Generative Modeling Without Language Pretraining
AI 导读
Halcyon 是一系列在 1 亿图文对数据集 GPIC 上完全从零预训练的早融合多模态自回归模型，证明强文生图系统无需互联网规模语言预训练。
正文 · 原文
Abstract: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 this https URL
| Comments: | 27 pages; Project website: this https URL |
| Subjects: | Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2610.26863 [cs.LG] |
| (or arXiv:2610.26863v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.26863 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Keshigeyan Chandrasegaran [view email]
[v1]
Mon, 19 Oct 2026 17:59:59 UTC (7,942 KB)
来源：HuggingFace Daily Papers · arxiv.org