Chipmunk：利用动态列稀疏增量加速扩散 Transformer
原标题：Chipmunk: Training-Free Acceleration of Diffusion Transformers with Dynamic Column-Sparse Deltas
AI 摘要
Together AI 和 UCSD 的研究人员提出了 Chipmunk，一种无需训练即可加速扩散 Transformer 推理的方法。该方法利用扩散步骤间激活值变化缓慢且稀疏的特性，通过缓存注意力权重和 MLP 激活，并动态计算稀疏增量，实现了高达 3.7 倍的视频生成加速（HunyuanVideo）和 1.6 倍的图像生成加速（FLUX.1-dev）。Chipmunk 还设计了硬件友好的稀疏模式，并开源了相关内核。
正文节选
TL;DR: We present Chipmunk, a training-free method to accelerate diffusion transformers with hardware-aware dynamic sparsity. Chipmunk caches attention weights and MLP activations from previous steps and dynamically computes a sparse “delta” against the cached weights. Chipmunk achieves up to 3.7x faster video generation on HunyuanVideo at 720x1280 resolution for a 5s video, and 1.6x faster image generations on FLUX.1-dev at 1280x768 resolution. This blog is cross-posted to the Sandy Research bl