Getting ViT in Shape: Scaling Laws for Compute-Optimal Model Design

Ibrahim Alabdulmohsin, Xiaohua Zhai, Alexander Kolesnikov, Lucas Beyer

Keywords: Vision transformer, scaling laws, compute-optimal model design, vision

TL;DR: We introduce new scaling laws for deriving compute-optimal model shapes and implement this in ViT. We demonstrate that scaled-down architectures trained at their optimal shapes for the right amount of compute, are comparable to fully-scaled models.

Abstract: Scaling laws have been recently employed to derive compute-optimal model size (number of parameters) for a given compute duration. We advance and refine such methods to infer compute-optimal model shapes, such as width and depth, and successfully implement this in vision transformers. Our shape-optimized vision transformer, SoViT, achieves results competitive with models that exceed twice its size, despite being pre-trained with an equivalent amount of compute. For example, SoViT-400m/14 achieves 90.3% fine-tuning accuracy on ILSRCV2012, surpassing the much larger ViT-g/14 and approaching ViT-G/14 under identical settings, with also less than half the inference cost. We conduct a thorough evaluation across multiple tasks, such as image classification, captioning, VQA and zero-shot transfer, demonstrating the effectiveness of our model across a broad range of domains and identifying limitations. Overall, our findings challenge the prevailing approach of blindly scaling up vision models and pave a path for a more informed scaling.

Submission Number: 7344