Pruning One More Token is Enough: Leveraging Latency-Workload Non-Linearities for Vision Transformers on the Edge
This work addresses the problem of high latency for vision transformers on edge devices, offering a training-free pruning method that is incremental but improves efficiency for specific deployment scenarios.
The paper tackles efficient deployment of vision transformers on edge devices by leveraging non-linear latency-workload relationships to prune tokens, reducing latency by 9-26% compared to other methods that increase it by 2-30%, and achieving 78.6%-84.5% ImageNet1K accuracy at similar latencies.
This paper investigates how to efficiently deploy vision transformers on edge devices for small workloads. Recent methods reduce the latency of transformer neural networks by removing or merging tokens, with small accuracy degradation. However, these methods are not designed with edge device deployment in mind: they do not leverage information about the latency-workload trends to improve efficiency. We address this shortcoming in our work. First, we identify factors that affect ViT latency-workload relationships. Second, we determine token pruning schedule by leveraging non-linear latency-workload relationships. Third, we demonstrate a training-free, token pruning method utilizing this schedule. We show other methods may increase latency by 2-30%, while we reduce latency by 9-26%. For similar latency (within 5.2% or 7ms) across devices we achieve 78.6%-84.5% ImageNet1K accuracy, while the state-of-the-art, Token Merging, achieves 45.8%-85.4%.