SANA-1.5通过三重优化，让文本生成图像模型在低算力下也能高效生成高质量图像。
SANA 1.5: Efficient Scaling of Training-Time and Inference-Time Compute in Linear Diffusion Transformer
- 采用深度增长与8比特优化器，训练参数从16亿增至48亿，算力需求大幅降低
- 通过块重要性分析实现任意大小压缩，质量损失极小，支持灵活部署
- 推理时重复采样提升模型表现，小模型可逼近大模型效果，适合资源受限场景
本文提出SANA-1.5，一种用于文本到图像生成的线性扩散Transformer，实现训练与推理阶段的高效扩展。在SANA-1.0基础上引入三项关键创新：(1) 高效训练扩展：采用深度增长范式，使模型参数从1.6B增至4.8B，结合内存高效的8比特优化器，显著降低计算开销；(2) 模型深度剪枝：提出块重要性分析技术，可将模型压缩至任意尺寸且质量损失极小；(3) 推理时扩展：设计重复采样策略，以增加计算代价换取模型容量，使小型模型在推理阶段达到大型模型质量。实验显示，SANA-1.5在GenEval上获得0.81的文本图像对齐得分，经VILA-Judge评估可进一步提升至0.96，刷新该基准新纪录。这些方法使模型能在不同算力预算下高效扩展，同时保持高质量输出。代码与预训练模型已开源。
原文摘要 · Abstract (English)
This paper presents SANA-1.5, a linear Diffusion Transformer for efficient scaling in text-to-image generation. Building upon SANA-1.0, we introduce three key innovations: (1) Efficient Training Scaling: A depth-growth paradigm that enables scaling from 1.6B to 4.8B parameters with significantly reduced computational resources, combined with a memory-efficient 8-bit optimizer. (2) Model Depth Pruning: A block importance analysis technique for efficient model compression to arbitrary sizes with minimal quality loss. (3) Inference-time Scaling: A repeated sampling strategy that trades computation for model capacity, enabling smaller models to match larger model quality at inference time. Through these strategies, SANA-1.5 achieves a text-image alignment score of 0.81 on GenEval, which can be further improved to 0.96 through inference scaling with VILA-Judge, establishing a new SoTA on GenEval benchmark. These innovations enable efficient model scaling across different compute budgets while maintaining high quality, making high-quality image generation more accessible. Our code and pre-trained models are released.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。