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

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Alabdulmohsin, Ibrahim, Zhai, Xiaohua, Kolesnikov, Alexander, Beyer, Lucas
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911750783238144
author Alabdulmohsin, Ibrahim
Zhai, Xiaohua
Kolesnikov, Alexander
Beyer, Lucas
author_facet Alabdulmohsin, Ibrahim
Zhai, Xiaohua
Kolesnikov, Alexander
Beyer, Lucas
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2305_13035
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Getting ViT in Shape: Scaling Laws for Compute-Optimal Model Design
Alabdulmohsin, Ibrahim
Zhai, Xiaohua
Kolesnikov, Alexander
Beyer, Lucas
Computer Vision and Pattern Recognition
Machine Learning
I.2.10; I.2.6
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.
title Getting ViT in Shape: Scaling Laws for Compute-Optimal Model Design
topic Computer Vision and Pattern Recognition
Machine Learning
I.2.10; I.2.6
url https://arxiv.org/abs/2305.13035