让一致性模型更稳定：用流匹配锚定训练，实现两步生成媲美千步效果
FACM: Flow-Anchored Consistency Models
- 用流匹配任务作为动态锚点，解决训练中速度场遗忘问题
- 2步生成时ImageNet上FID达1.32，1步生成时为1.70，性能领先
- 适用于追求高效生成的图像生成研究者，尤其适合大模型部署
连续时间一致性模型（CMs）虽能实现少步高效生成，但面临训练不稳定的难题。我们指出，其根源在于仅优化快捷目标导致瞬时速度场被灾难性遗忘。为此提出流动锚定一致性模型（FACM），以流匹配（FM）任务作为主目标的动态锚点，确保训练中轨迹保真度。核心是创新的扩展时间区间策略，统一单模型优化并解耦两任务，实现稳定且与架构无关的训练。通过蒸馏预训练LightningDiT模型，本方法在ImageNet 256x256上实现2步生成（NFE=2）FID 1.32，1步生成（NFE=1）FID 1.70，达当前最优。为解决可扩展性问题，设计内存高效的Chain-JVP，克服与FSDP的关键冲突，使140亿参数模型（Wan 2.2）训练成为可能，将文本到图像推理步骤从2×40降至2-8步。代码与预训练模型见：https://github.com/ali-vilab/FACM。
原文摘要 · Abstract (English)
Continuous-time Consistency Models (CMs) promise efficient few-step generation but face significant challenges with training instability. We argue this instability stems from a fundamental conflict: Training the network exclusively on a shortcut objective leads to the catastrophic forgetting of the instantaneous velocity field that defines the flow. Our solution is to explicitly anchor the model in the underlying flow, ensuring high trajectory fidelity during training. We introduce the Flow-Anchored Consistency Model (FACM), where a Flow Matching (FM) task serves as a dynamic anchor for the primary CM shortcut objective. Key to this Flow-Anchoring approach is a novel expanded time interval strategy that unifies optimization for a single model while decoupling the two tasks to ensure stable, architecturally-agnostic training. By distilling a pre-trained LightningDiT model, our method achieves a state-of-the-art FID of 1.32 with two steps (NFE=2) and 1.70 with just one step (NFE=1) on ImageNet 256x256. To address the challenge of scalability, we develop a memory-efficient Chain-JVP that resolves key incompatibilities with FSDP. This method allows us to scale FACM training on a 14B parameter model (Wan 2.2), accelerating its Text-to-Image inference from 2x40 to 2-8 steps. Our code and pretrained models: https://github.com/ali-vilab/FACM.
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。