研究人员推出 SoftCap，这是一种新颖的无需训练的控制层，旨在加速扩散 Transformer (DiTs)。该方法通过智能管理昂贵的完整 Transformer 评估的执行来优化推理过程。SoftCap 采用轨迹漂移观察器来评估缓存风险，并使用软预算 PI 控制器动态调整完整步骤的阈值，从而在不牺牲视觉质量的情况下提高效率。 AI
影响 SoftCap 为更高效的 DiT 推理提供了一条途径，有可能降低计算成本并加快图像生成过程。
排序理由 该集群包含一篇详细介绍加速 AI 模型新方法的学术论文。
在 Hugging Face Daily Papers 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
Diffusion Transformers (DiTs) achieve strong visual quality, but their iterative denoising process requires many costly Transformer evaluations. Training-free acceleration methods reduce this cost by caching, forecasting, or verifying intermediate features, yet the runtime decisi…
arXiv:2605.27075v1 Announce Type: new Abstract: Diffusion Transformers (DiTs) achieve strong visual quality, but their iterative denoising process requires many costly Transformer evaluations. Training-free acceleration methods reduce this cost by caching, forecasting, or verifyi…
Diffusion Transformers (DiTs) achieve strong visual quality, but their iterative denoising process requires many costly Transformer evaluations. Training-free acceleration methods reduce this cost by caching, forecasting, or verifying intermediate features, yet the runtime decisi…