TAP：面向无训练扩散加速的词元自适应预测器框架
TAP: A Token-Adaptive Predictor Framework for Training-Free Diffusion Acceleration
把「逐词元自适应选择预测器」作为即插即用的无训练加速模块，可直接套到已有扩散/流匹配采样器上，用首层单次前向作为低成本探针来分配不同阶数/视野的泰勒外推，替代全局固定缓存策略。
原文依据 TAP uses a single full evaluation of the model's first layer as a low-cost probe to compute proxy losses for a compact family of candidate predictors
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具有强大的生成性能，但由于需要反复进行全模型去噪前向，推理速度仍然很慢。我们提出词元自适应预测器（TAP），这是一个无需训练、由探针驱动的框架，在每个采样步为每个词元自适应地选择一个预测器。TAP 将模型首层的一次完整计算作为低成本探针，为一个小型候选预测器集合（主要以不同阶数与视野的泰勒展开实现）计算代理损失，然后把代理误差最小的预测器分配给每个词元。这种逐词元的「先探测再选择」策略利用了异质的时序动态，无需额外训练，并且兼容多种预测器设计。TAP 带来的额外开销可以忽略不计，同时能在几乎不损失或完全不损失感知质量的前提下实现大幅加速。在多种扩散架构与生成任务上的大量实验表明，与固定的全局预测器和仅依赖缓存的基线相比，TAP 显著推进了精度与效率之间的前沿。
英文原文摘要 1086 字符
Diffusion models achieve strong generative performance but remain slow at inference due to the need for repeated full-model denoising passes. We present Token-Adaptive Predictor (TAP), a training-free, probe-driven framework that adaptively selects a predictor for each token at every sampling step. TAP uses a single full evaluation of the model's first layer as a low-cost probe to compute proxy losses for a compact family of candidate predictors (instantiated primarily with Taylor expansions of varying order and horizon), then assigns each token the predictor with the smallest proxy error. This per-token "probe-then-select" strategy exploits heterogeneous temporal dynamics, requires no additional training, and is compatible with various predictor designs. TAP incurs negligible overhead while enabling large speedups with little or no perceptual quality loss. Extensive experiments across multiple diffusion architectures and generation tasks show that TAP substantially improves the accuracy-efficiency frontier compared to fixed global predictors and caching-only baselines.