Otil：通过通信高效的多 GPU 并行加速扩散模型推理
Otil: Accelerating Diffusion Model Inference via Communication-Efficient Multi-GPU Parallelism
对做扩散模型高分辨率/视频生成的研究者，这是一种即插即用的多卡并行推理方案：利用相邻去噪步潜变量变化微小的性质，只同步信息量最大的潜变量子块并周期性轮询全部空间区域，可在不改架构、不重训的前提下把 GPU 间通信降低最多 87.5%，与少步采样器和低秩适配叠加后进一步提速。
原文依据 Otil identifies and synchronizes only the most informative latent sub-blocks and introduces a dynamic polling mechanism
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论扩散模型（DMs）近期在多种模态上取得了显著成功，包括高保真图像与视频合成。然而，其固有的逐步顺序去噪过程会带来大量累积延迟，显著降低用户体验。现有的多 GPU 并行方法虽能缓解延迟，却往往产生高昂的 GPU 间通信开销，抵消了大部分性能收益。我们提出 Otil（Only Transmit Informative Latents，仅传输信息性潜变量），一个通信高效的并行框架，用于加速扩散推理。Otil 能够在保持生成质量的同时最小化 GPU 间的冗余数据交换。我们的关键洞见是：在连续的去噪过程之间，潜变量激活仅发生微小变化。利用这一性质，Otil 只识别并同步信息量最大的潜变量子块，并引入一种动态轮询机制，周期性地重新访问所有空间区域，从而在无需多余通信的情况下确保完整覆盖。该框架完全即插即用，并与快速采样与架构加速算法保持兼容，无需任何重训练或架构修改。与当前最优的并行方法相比，Otil 将 GPU 间通信最多降低 87.5%，在两张 GPU 上使用 Stable Diffusion v1.5 取得 1.8 倍加速，在四张 GPU 上使用 Stable Diffusion XL 取得 2.6 倍加速。当与少步采样器（30 步）和低秩适配模型结合时，在 2 张 GPU 上加速进一步提升至 2.46 倍至 2.84 倍。这些结果展示了 Otil 在保持生成保真度的同时，实现可扩展且高效的多 GPU 扩散推理的巨大潜力。
英文原文摘要 1695 字符
Diffusion models (DMs) have recently achieved remarkable success across diverse modalities, including high-fidelity image and video synthesis.However, their inherent step sequential denoising process introduces substantial cumulative latency, which significantly degrades user experience. While existing multi-GPU parallelization motheds can alleviate latency, they often incur prohibitive GPU-GPU communication overhead, offsetting much of the performance gain.We present Otil (Only Transmit Informative Latents), a communication-efficient parallel framework for accelerating diffusion inference.% Otil can minimizes redundant data exchange across GPUs while preserving generation quality.Our key insight is that latent activations change only marginally between consecutive denoising process. Leveraging this property, Otil identifies and synchronizes only the most informative latent sub-blocks and introduces a dynamic polling mechanism that periodically revisits all spatial regions, ensuring complete coverage without unnecessary communication. The framework is fully plug-and-play and remains compatible with fast sample and architectural acceleration algorithms, without requiring any retraining or architectural modification.Otil reduces GPU-GPU communication up to 87.5% compared with SOTA parallelism methods, achieving 1.8xspeedup on two GPUs with Stable Diffusion v1.5 and 2.6xon four GPUs with Stable Diffusion XL. When combined with few-step samplers (30 steps) and LoRA models, the acceleration further increases to 2.46x-2.84xon 2 GPUs. These demonstrate the strong potential of Otil for scalable and efficient multi-GPU diffusion inference while preserving generation fidelity.