Model card for CLIP ViT-B-32 256x256 trained DataComp-1B

Table of Contents

1. Model Details
2. Uses
3. Training Details
4. Evaluation
5. How To Get Started With the Model
6. Acknowledgements
7. Citation

Model Details

Model Description

A CLIP ViT-B/32 model trained with the DataComp-1B (https://github.com/mlfoundations/datacomp) using OpenCLIP (https://github.com/mlfoundations/open_clip) at 256x256 resolution.

Model training done on the JURECA cluster by Mehdi Cherti.

Uses

As per the original OpenAI CLIP model card, this model is intended as a research output for research communities. We hope that this model will enable researchers to better understand and explore zero-shot, arbitrary image classification. We also hope it can be used for interdisciplinary studies of the potential impact of such model.

The OpenAI CLIP paper includes a discussion of potential downstream impacts to provide an example for this sort of analysis. Additionally, the DataComp paper (https://arxiv.org/abs/2304.14108) include additional discussion as it relates specifically to the training dataset.

Direct Use

Zero-shot image classification, image and text retrieval, among others.

Downstream Use

Image classification and other image task fine-tuning, linear probe image classification, image generation guiding and conditioning, among others.

Out-of-Scope Use

As per the OpenAI models,

Any deployed use case of the model - whether commercial or not - is currently out of scope. Non-deployed use cases such as image search in a constrained environment, are also not recommended unless there is thorough in-domain testing of the model with a specific, fixed class taxonomy. This is because our safety assessment demonstrated a high need for task specific testing especially given the variability of CLIP’s performance with different class taxonomies. This makes untested and unconstrained deployment of the model in any use case currently potentially harmful.

Certain use cases which would fall under the domain of surveillance and facial recognition are always out-of-scope regardless of performance of the model. This is because the use of artificial intelligence for tasks such as these can be premature currently given the lack of testing norms and checks to ensure its fair use.

Training Details

Training Data

This model was trained with the 1.4 Billion samples of the DataComp-1B dataset (https://arxiv.org/abs/2304.14108).

IMPORTANT NOTE: The motivation behind dataset creation is to democratize research and experimentation around large-scale multi-modal model training and handling of uncurated, large-scale datasets crawled from publically available internet. Our recommendation is therefore to use the dataset for research purposes. Be aware that this large-scale dataset is uncurated. Keep in mind that the uncurated nature of the dataset means that collected links may lead to strongly discomforting and disturbing content for a human viewer. Therefore, please use the demo links with caution and at your own risk. It is possible to extract a “safe” subset by filtering out samples based on the safety tags (using a customized trained NSFW classifier that we built). While this strongly reduces the chance for encountering potentially harmful content when viewing, we cannot entirely exclude the possibility for harmful content being still present in safe mode, so that the warning holds also there. We think that providing the dataset openly to broad research and other interested communities will allow for transparent investigation of benefits that come along with training large-scale models as well as pitfalls and dangers that may stay unreported or unnoticed when working with closed large datasets that remain restricted to a small community. Providing our dataset openly, we however do not recommend using it for creating ready-to-go industrial products, as the basic research about general properties and safety of such large-scale models, which we would like to encourage with this release, is still in progress.

SLURM script

#!/bin/bash -x
#SBATCH --nodes=24
#SBATCH --gres=gpu:4
#SBATCH --ntasks-per-node=4
#SBATCH --cpus-per-task=12
#SBATCH --time=24:00:00
source /path/miniconda/bin/activate
export CUDA_VISIBLE_DEVICES=0,1,2,3
export MASTER_PORT=12802
master_addr=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
export MASTER_ADDR=$master_addr"i"
echo "MASTER_ADDR="$MASTER_ADDR
srun --cpu-bind=v --cpus-per-task=12 python -u -m training.main --aug-cfg scale='(0.4, 1.0)' color_jitter='(0.32, 0.32, 0.32, 0.08)' color_jitter_prob=0.8 gray_scale_prob=0.2 use_timm=True \
--save-frequency 1 \
--zeroshot-frequency 1 \
--dataset-type webdataset \
--train-data '/path/to/data' \
--report-to tensorboard \
--train-num-samples 1398270000 \
--warmup 2000 \
--batch-size 896 \
--epochs 24 \
--workers 8 \
--model ViT-B-32-256 \
--logs logs \
--seed 0 \
--ddp-static-graph \
--local-loss \
--gather-with-grad \
--lr 0.001 \
--log-every-n-steps 20 \
--save-most-recent \
--resume latest \
--grad-checkpointing \
--precision amp_bfloat16 \
--beta1 0.9 \
--beta2 0.95 \
--wd 0.2

Evaluation

Evaluation done on 38 datasets, using LAION CLIP Benchmark.

Testing Data, Factors & Metrics

Testing Data

The testing is performed on a suite of 38 datasets. See our paper for more details (https://arxiv.org/abs/2304.14108).

Results

The model achieves a 72.7% zero-shot top-1 accuracy on ImageNet-1k, 64.4% image retrieval recall@5 and 80.7% text retrieval recall@5 on COCO captions.

How to Get Started with the Model

Zero-shot classification example:

import torch
from PIL import Image
import open_clip

model, _, preprocess = open_clip.create_model_and_transforms('hf-hub:laion/CLIP-ViT-B-32-256x256-DataComp-s34B-b86K')
model.eval()  # model in train mode by default, impacts some models with BatchNorm or stochastic depth active
tokenizer = open_clip.get_tokenizer('hf-hub:laion/CLIP-ViT-B-32-256x256-DataComp-s34B-b86K')

image = preprocess(Image.open("docs/CLIP.png")).unsqueeze(0)
text = tokenizer(["a diagram", "a dog", "a cat"])

with torch.no_grad(), torch.autocast("cuda"):
    image_features = model.encode_image(image)
    text_features = model.encode_text(text)
    image_features /= image_features.norm(dim=-1, keepdim=True)
    text_features /= text_features.norm(dim=-1, keepdim=True)

    text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1)

print("Label probs:", text_probs)  # prints: [[1., 0., 0.]]

Acknowledgements

We gratefully acknowledge the computing time granted by the John von Neumann Institute for Computing (NIC) and provided on the supercomputer JURECA at Jülich Supercomputing Centre (JSC).

Citation

BibTeX:

DataComp

@article{datacomp,
  title={DataComp: In search of the next generation of multimodal datasets},
  author={Samir Yitzhak Gadre, Gabriel Ilharco, Alex Fang, Jonathan Hayase, Georgios Smyrnis, Thao Nguyen, Ryan Marten, Mitchell Wortsman, Dhruba Ghosh, Jieyu Zhang, Eyal Orgad, Rahim Entezari, Giannis Daras, Sarah Pratt, Vivek Ramanujan, Yonatan Bitton, Kalyani Marathe, Stephen Mussmann, Richard Vencu, Mehdi Cherti, Ranjay Krishna, Pang Wei Koh, Olga Saukh, Alexander Ratner, Shuran Song, Hannaneh Hajishirzi, Ali Farhadi, Romain Beaumont, Sewoong Oh, Alex Dimakis, Jenia Jitsev, Yair Carmon, Vaishaal Shankar, Ludwig Schmidt},
  journal={arXiv preprint arXiv:2304.14108},
  year={2023}
}

OpenAI CLIP paper

@inproceedings{Radford2021LearningTV,
  title={Learning Transferable Visual Models From Natural Language Supervision},
  author={Alec Radford and Jong Wook Kim and Chris Hallacy and A. Ramesh and Gabriel Goh and Sandhini Agarwal and Girish Sastry and Amanda Askell and Pamela Mishkin and Jack Clark and Gretchen Krueger and Ilya Sutskever},
  booktitle={ICML},
  year={2021}
}

OpenCLIP software

@software{ilharco_gabriel_2021_5143773,
  author       = {Ilharco, Gabriel and
                  Wortsman, Mitchell and
                  Wightman, Ross and
                  Gordon, Cade and
                  Carlini, Nicholas and
                  Taori, Rohan and
                  Dave, Achal and
                  Shankar, Vaishaal and
                  Namkoong, Hongseok and
                  Miller, John and
                  Hajishirzi, Hannaneh and
                  Farhadi, Ali and
                  Schmidt, Ludwig},
  title        = {OpenCLIP},
  month        = jul,
  year         = 2021,
  note         = {If you use this software, please cite it as below.},
  publisher    = {Zenodo},
  version      = {0.1},
  doi          = {10.5281/zenodo.5143773},
  url          = {https://doi.org/10.5281/zenodo.5143773}
}

CLIP benchmark software

@software{cherti_2025_15403103,
  author       = {Cherti, Mehdi and
                  Beaumont, Romain},
  title        = {CLIP benchmark},
  month        = may,
  year         = 2025,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.15403103},
  url          = {https://doi.org/10.5281/zenodo.15403103},
  swhid        = {swh:1:dir:8cf49a5dd06f59224844a1e767337a1d14ee56c2
                   ;origin=https://doi.org/10.5281/zenodo.15403102;vi
                   sit=swh:1:snp:dd153b26f702d614346bf814f723d59fef3d
                   77a2;anchor=swh:1:rel:cff2aeb98f42583b44fdab5374e9
                   fa71793f2cff;path=CLIP\\_benchmark-main
                  },
}