# An Inverse Scaling Law for CLIP Training

arXiv: https://arxiv.org/abs/2305.07017 · HTML: https://mlsanity.com/paper/2305.07017/

## Key facts

- Cited by 77 papers in the map (data through 14 Oct 2026).
- Builds on 66 of its 73 references that are in the map.
- First posted on arXiv 11 May 2023.

## Builds on (most cited first, 30 of 66)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [High-Resolution Image Synthesis with Latent Diffusion Models](https://mlsanity.com/paper/2112.10752/) (2021)
- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://mlsanity.com/paper/1910.10683/) (2019)
- [A Simple Framework for Contrastive Learning of Visual Representations](https://mlsanity.com/paper/2002.05709/) (2020)
- [Masked Autoencoders Are Scalable Vision Learners](https://mlsanity.com/paper/2111.06377/) (2021)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding](https://mlsanity.com/paper/2205.11487/) (2022)
- [PaLM: Scaling Language Modeling with Pathways](https://mlsanity.com/paper/2204.02311/) (2022)
- [SGDR: Stochastic Gradient Descent with Warm Restarts](https://mlsanity.com/paper/1608.03983/) (2016)
- [Flamingo: a Visual Language Model for Few-Shot Learning](https://mlsanity.com/paper/2204.14198/) (2022)
- [LAION-5B: An open large-scale dataset for training next generation image-text models](https://mlsanity.com/paper/2210.08402/) (2022)
- [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://mlsanity.com/paper/2102.05918/) (2021)
- [A ConvNet for the 2020s](https://mlsanity.com/paper/2201.03545/) (2022)
- [Scaling Instruction-Finetuned Language Models](https://mlsanity.com/paper/2210.11416/) (2022)
- [The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization](https://mlsanity.com/paper/2006.16241/) (2020)
- [LAION-400M: Open Dataset of CLIP-Filtered 400 Million Image-Text Pairs](https://mlsanity.com/paper/2111.02114/) (2021)
- [Do ImageNet Classifiers Generalize to ImageNet?](https://mlsanity.com/paper/1902.10811/) (2019)
- [Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour](https://mlsanity.com/paper/1706.02677/) (2017)
- [Natural Adversarial Examples](https://mlsanity.com/paper/1907.07174/) (2019)
- [Big Self-Supervised Models are Strong Semi-Supervised Learners](https://mlsanity.com/paper/2006.10029/) (2020)
- [CoCa: Contrastive Captioners are Image-Text Foundation Models](https://mlsanity.com/paper/2205.01917/) (2022)
- [Conceptual 12M: Pushing Web-Scale Image-Text Pre-Training To Recognize Long-Tail Visual Concepts](https://mlsanity.com/paper/2102.08981/) (2021)

## Built on by (most cited first, 30 of 77)

- [Perception Encoder: The best visual embeddings are not at the output of the network](https://mlsanity.com/paper/2504.13181/) (2025)
- [Pixtral 12B](https://mlsanity.com/paper/2410.07073/) (2024)
- [SAM-CLIP: Merging Vision Foundation Models towards Semantic and Spatial Understanding](https://mlsanity.com/paper/2310.15308/) (2023)
- [Foundational Models Defining a New Era in Vision: A Survey and Outlook](https://mlsanity.com/paper/2307.13721/) (2023)
- [A Picture is Worth More Than 77 Text Tokens: Evaluating CLIP-Style Models on Dense Captions](https://mlsanity.com/paper/2312.08578/) (2023)
- [MobileCLIP: Fast Image-Text Models through Multi-Modal Reinforced Training](https://mlsanity.com/paper/2311.17049/) (2023)
- [What If We Recaption Billions of Web Images with LLaMA-3?](https://mlsanity.com/paper/2406.08478/) (2024)
- [RewardDance: Reward Scaling in Visual Generation](https://mlsanity.com/paper/2509.08826/) (2025)
- [Two Effects, One Trigger: On the Modality Gap, Object Bias, and Information Imbalance in Contrastive Vision-Language Models](https://mlsanity.com/paper/2404.07983/) (2024)
- [MLLMs are Deeply Affected by Modality Bias](https://mlsanity.com/paper/2505.18657/) (2025)
- [ViTamin: Designing Scalable Vision Models in the Vision-Language Era](https://mlsanity.com/paper/2404.02132/) (2024)
- [SaulLM-54B & SaulLM-141B: Scaling Up Domain Adaptation for the Legal Domain](https://mlsanity.com/paper/2407.19584/) (2024)
- [Procedure-Aware Surgical Video-language Pretraining with Hierarchical Knowledge Augmentation](https://mlsanity.com/paper/2410.00263/) (2024)
- [Effective pruning of web-scale datasets based on complexity of concept clusters](https://mlsanity.com/paper/2401.04578/) (2024)
- [Revisiting Adversarial Training at Scale](https://mlsanity.com/paper/2401.04727/) (2024)
- [Data curation via joint example selection further accelerates multimodal learning](https://mlsanity.com/paper/2406.17711/) (2024)
- [UniBench: Visual Reasoning Requires Rethinking Vision-Language Beyond Scaling](https://mlsanity.com/paper/2408.04810/) (2024)
- [The Scalability of Simplicity: Empirical Analysis of Vision-Language Learning with a Single Transformer](https://mlsanity.com/paper/2504.10462/) (2025)
- [Dataset Decomposition: Faster LLM Training with Variable Sequence Length Curriculum](https://mlsanity.com/paper/2405.13226/) (2024)
- [CLIPA-v2: Scaling CLIP Training with 81.1% Zero-shot ImageNet Accuracy within a 10,000 Budget; An Extra 4,000 Unlocks 81.8% Accuracy](https://mlsanity.com/paper/2306.15658/) (2023)
- [MoDE: CLIP Data Experts via Clustering](https://mlsanity.com/paper/2404.16030/) (2024)
- [MobileCLIP2: Improving Multi-Modal Reinforced Training](https://mlsanity.com/paper/2508.20691/) (2025)
- [An Empirical Study Into What Matters for Calibrating Vision-Language Models](https://mlsanity.com/paper/2402.07417/) (2024)
- [X-Transfer Attacks: Towards Super Transferable Adversarial Attacks on CLIP](https://mlsanity.com/paper/2505.05528/) (2025)
- [Rejuvenating image-GPT as Strong Visual Representation Learners](https://mlsanity.com/paper/2312.02147/) (2023)
- [OpenVision: A Fully-Open, Cost-Effective Family of Advanced Vision Encoders for Multimodal Learning](https://mlsanity.com/paper/2505.04601/) (2025)
- [Mind the Modality Gap: Towards a Remote Sensing Vision-Language Model via Cross-modal Alignment](https://mlsanity.com/paper/2402.09816/) (2024)
- [If CLIP Could Talk: Understanding Vision-Language Model Representations Through Their Preferred Concept Descriptions](https://mlsanity.com/paper/2403.16442/) (2024)
- [SCAM: A Real-World Typographic Robustness Evaluation for Multimodal Foundation Models](https://mlsanity.com/paper/2504.04893/) (2025)
- [Locality Alignment Improves Vision-Language Models](https://mlsanity.com/paper/2410.11087/) (2024)

## Abstract (arXiv)

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.