English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

Token merging has emerged as a new paradigm that can accelerate the inferenceof Vision Transformers (ViTs) without any retraining or fine-tuning. To pushthe frontier of training-free acceleration in ViTs, we improve token merging byadding the perspectives of 1) activation outliers and 2) hierarchicalrepresentations. Through a careful analysis of the attention behavior in ViTs,we characterize a delayed onset of the convergent attention phenomenon, whichmakes token merging undesirable in the bottom blocks of ViTs. Moreover, weaugment token merging with a hierarchical processing scheme to capturemulti-scale redundancy between visual tokens. Combining these two insights, webuild a unified inference framework called DSM: Delayed Spatial Merging. Weextensively evaluate DSM on various ViT model scales (Tiny to Huge) and tasks(ImageNet-1k and transfer learning), achieving up to 1.8$\times$ FLOP reductionand 1.6$\times$ throughput speedup at a negligible loss while being two ordersof magnitude faster than existing methods.

Training-Free Acceleration of ViTs with Delayed Spatial Merging