Attention Is All You Need For Mixture-of-Depths Routing

TL;DR

This paper introduces A-MoD, an attention-based routing mechanism for Mixture-of-Depths models that improves training efficiency and accuracy without additional parameters, enhancing large model deployment.

Contribution

A-MoD leverages existing attention maps for routing, eliminating extra trainable layers and simplifying training of MoD models.

Findings

Up to 2% higher accuracy on ImageNet

2x faster transfer learning

Simplifies training without extra parameters

Abstract

Advancements in deep learning are driven by training models with increasingly larger numbers of parameters, which in turn heightens the computational demands. To address this issue, Mixture-of-Depths (MoD) models have been proposed to dynamically assign computations only to the most relevant parts of the inputs, thereby enabling the deployment of large-parameter models with high efficiency during inference and training. These MoD models utilize a routing mechanism to determine which tokens should be processed by a layer, or skipped. However, conventional MoD models employ additional network layers specifically for the routing which are difficult to train, and add complexity and deployment overhead to the model. In this paper, we introduce a novel attention-based routing mechanism A-MoD that leverages the existing attention map of the preceding layer for routing decisions within the current layer. Compared to standard routing, A-MoD allows for more efficient training as it introduces no additional trainable parameters and can be easily adapted from pretrained transformer models. Furthermore, it can increase the performance of the MoD model. For instance, we observe up to 2% higher accuracy on ImageNet compared to standard routing and isoFLOP ViT baselines. Furthermore, A-MoD improves the MoD training convergence, leading to up to 2x faster transfer learning.