自强制：弥合自回归视频扩散中的训练-测试差距
Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion
AI总结：
本文提出 Self Forcing 训练范式，通过在训练时基于自生成输出进行自回归展开和视频级整体损失监督，并配合少步扩散、梯度截断与滚动 KV 缓存，在单 GPU 上实现亚秒级实时视频生成，弥合自回归视频扩散中的训练-测试差距。
摘要：
我们提出 Self Forcing，一种用于自回归视频扩散模型的新型训练范式。它解决了长期存在的曝光偏差问题，即模型在真实上下文上训练，但在推理时必须基于自身不完美的输出生成序列。与先前基于真实上下文帧对后续帧进行去噪的方法不同，Self Forcing 通过在训练期间使用键值（KV）缓存执行自回归展开，使每一帧的生成都以先前自生成输出为条件。该策略通过视频级别的整体损失进行监督，直接评估整个生成序列的质量，而不是仅仅依赖传统的逐帧目标。为确保训练效率，我们采用少步扩散模型以及随机梯度截断策略，有效平衡计算成本与性能。我们进一步引入滚动 KV 缓存机制，实现高效的自回归视频外推。大量实验表明，我们的方法在单个 GPU 上实现了亚秒级延迟的实时流式视频生成，同时达到甚至超越显著更慢且非因果的扩散模型的生成质量。项目网站：http://self-forcing.github.io/
英文摘要：
We introduce Self Forcing, a novel training paradigm for autoregressive video diffusion models. It addresses the longstanding issue of exposure bias, where models trained on ground-truth context must generate sequences conditioned on their own imperfect outputs during inference. Unlike prior methods that denoise future frames based on ground-truth context frames, Self Forcing conditions each frame's generation on previously self-generated outputs by performing autoregressive rollout with key-value (KV) caching during training. This strategy enables supervision through a holistic loss at the video level that directly evaluates the quality of the entire generated sequence, rather than relying solely on traditional frame-wise objectives. To ensure training efficiency, we employ a few-step diffusion model along with a stochastic gradient truncation strategy, effectively balancing computational cost and performance. We further introduce a rolling KV cache mechanism that enables efficient autoregressive video extrapolation. Extensive experiments demonstrate that our approach achieves real-time streaming video generation with sub-second latency on a single GPU, while matching or even surpassing the generation quality of significantly slower and non-causal diffusion models. Project website: http://self-forcing.github.io/
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
