Pith. sign in

REVIEW 2 cited by

An Inverse Scaling Law for CLIP Training

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2305.07017 v2 pith:3RV3KM22 submitted 2023-05-11 cs.CV

An Inverse Scaling Law for CLIP Training

classification cs.CV
keywords cliptrainingscalingtextdaysimageassociatedbarrier
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

CLIP, one of the pioneering foundation models that connect images and text, has enabled many recent breakthroughs in computer vision. However, its associated training cost is prohibitively high, imposing a significant barrier to its widespread exploration. In this paper, we present a surprising finding that there exists an inverse scaling law for CLIP training, whereby the larger the image/text encoders used, the shorter the sequence length of image/text tokens that can be applied in training. Moreover, we showcase that the strategy for reducing image/text token length plays a crucial role in determining the quality of this scaling law. As a result of this finding, we are able to successfully train CLIP even with limited computational resources. For example, using 8 A100 GPUs, our CLIP models achieve zero-shot top-1 ImageNet-1k accuracies of 63.2% in ~2 days, 67.8% in ~3 days, and 69.3% in ~4 days. Our method also works well when scaling up -- with G/14, we register a new record of 83.0% ImageNet-1k zero-shot accuracy, and meanwhile accelerate the training by ~33x compared to its OpenCLIP counterpart. By reducing the computation barrier associated with CLIP, we hope to inspire more research in this field, particularly from academics. Our code is available at https://github.com/UCSC-VLAA/CLIPA.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Understanding Space Is Rocket Science -- Only Top Reasoning Models Can Solve Spatial Understanding Tasks

    cs.CV 2025-09 conditional novelty 6.0

    A new contrastive real-image benchmark shows most vision-language models fail spatial relation tasks, while chain-of-thought reasoning models approach human-level accuracy.

  2. MobileCLIP2: Improving Multi-Modal Reinforced Training

    cs.CV 2025-08 conditional novelty 5.0

    MobileCLIP2 combines DFN-trained teachers, a fine-tuned CoCa captioner, and new 5-stage FastViT variants to set state-of-the-art ImageNet-1k zero-shot accuracy at low latency.