BLOG
Announcing DataComp: In search of the next generation of multimodal datasets
by: Gabriel Ilharco, 27 Apr, 2023
[ Paper ] [ Code ] [ Website ] About a year ago, we released LAION-5B, a billion-scale open-source image-text dataset. Since then, LAION-5B has become a staple in the open-source machine learning ecosystem, powering open-source models like OpenCLIP, OpenFlamingo, and Stable Diffusion. From the begin...
A new Paella: Simple & Efficient Text-To-Image generation
by: Dominic Rampas and Pablo Pernias, 15 Apr, 2023
Overview. We are releasing a new Paella model which builds on top of our initial paper https://arxiv.org/abs/2211.07292. Paella is a text-to-image model that works in a quantized latent space and learns similarly to MUSE and Diffusion models. Paella is similar to MUSE as it also works on discrete t...
Petition for keeping up the progress tempo on AI research while securing its transparency and safety.
by: LAION.ai, 29 Mar, 2023
LINK TO OUR PETITION Authors: Christoph Schuhmann, Huu Nguyen, Robert Kaczmarczyk, Jenia Jitsev & LAION community Securing Our Digital Future: Calling for CERN like international organization to transparently coordinate and progress on large-scale AI research and its safety In an era of unparall...
Announcing OpenFlamingo: An open-source framework for training vision-language models with in-context learning
by: Anas Awadalla and Irena Gao, 28 Mar, 2023
Overview. We are thrilled to announce the release of OpenFlamingo, an open-source reproduction of DeepMind's Flamingo model. At its core, OpenFlamingo is a framework that enables training and evaluation of large multimodal models (LMMs). Check out our GitHub repository and demo to get started! For t...
The OIG Dataset
by: By Huu Nguyen - Ontocord.ai, Sameer Suri, Ken Tsui , Shahules786, Together.xyz team, and Christoph Schuhmann - LAION.ai, 10 Mar, 2023
The Open Instruction Generalist (OIG) dataset is a large open source instruction dataset that currently contains ~43M instructions. OIG is one of many chatbot datasets that LAION, along with its volunteers, Ontocord, Together and other members of the open source community, will be releasing and is i...
Training Contrastive Captioners
by: Giovanni Puccetti, Maciej Kilian, Romain Beaumont, 02 Feb, 2023
We introduce a new model type to OpenClip Contrastive Captioners (CoCa) [1]. This model adds an autoregressive objective (generation) on top of the CLIP contrastive one. The architecture is composed of three parts, the first two are similar to those composing a CLIP model and the third is a text dec...
Clip-Retrieval Update: H-14 Index & SLURM Inference
by: no usr, 31 Jan, 2023
Today we release a KNN index for LAION-5B that allows for fast queries of the dataset with the open clip ViT-H-14 CLIP model. This means that users can search through billions of samples quickly and easily, making it a powerful tool for various applications such as image and text retrieval, data fil...
Reaching 80% zero-shot accuracy with OpenCLIP: ViT-G/14 trained on LAION-2B
by: Mitchell Wortsman, 24 Jan, 2023
We have trained a new ViT-G/14 CLIP model with OpenCLIP which achieves 80.1% zero-shot accuracy on ImageNet and 74.9% zero-shot image retrieval (Recall@5) on MS COCO. As of January 2023, this is the best open source CLIP model. We believe this is interesting because: CLIP models are useful for zero...
Collaboration between LAION and the Stable Horde
by: Konstantinos Thoukydidis, hlky, 08 Jan, 2023
Author: Konstantinos Thoukydidis, hlky We are happy to announce that LAION will be assisted by the Stable Horde to provide aesthetic ratings for existing datasets and a completely new dataset of Stable Diffusion generations, which will also be rated by their community. We wrote in the past about LAI...
Laion coco: 600M synthetic captions from Laion2B-en
by: Christoph Schuhmann, Andreas Köpf, Richard Vencu, Theo Coombes, Romain Beaumont, 15 Sep, 2022
Author: Christoph Schuhmann, Andreas Köpf , Theo Coombes, Richard Vencu, Benjamin Trom , Romain Beaumont We present LAION-COCO, the world’s largest dataset of 600M generated high-quality captions for publicly available web-images Laion5B has five billion natural captions. They provide a lot of infor...
Laion translated: 3B captions translated to English from laion5B
by: Marianna Nezhurina, Romain Beaumont, Richard Vencu and Christoph Schuhmann, 15 Sep, 2022
Author: Marianna Nezhurina Romain Beaumont Richard Vencu Christoph Schuhmann Laion5B dataset was automatically collected from a section of the human web (common crawl). Can models generate different and interesting data compared to what humans write? That’s a question we are interested in investigat...
Large scale openCLIP: L/14, H/14 and g/14 trained on LAION-2B
by: Romain Beaumont, 15 Sep, 2022
We trained three large CLIP models with OpenCLIP: ViT-L/14, ViT-H/14 and ViT-g/14 (ViT-g/14 was trained only for about a third the epochs compared to the rest). The H/14 model achieves 78.0% zero shot top-1 accuracy on ImageNet and 73.4% on zero-shot image retrieval at Recall@5 on MS COCO. As of Sep...
LAION-Aesthetics
by: Christoph Schuhmann, 16 Aug, 2022
We present LAION-Aesthetics, several collections of subsets from LAION 5B with high visual quality. To create LAION-Aesthetics we trained several lightweight models that predict the rating people gave when they were asked “How much do you like this image on a scale from 1 to 10?”. LAION-Aesthetics ...
LAION-5B: A NEW ERA OF OPEN LARGE-SCALE MULTI-MODAL DATASETS
by: Romain Beaumont, 31 Mar, 2022
We present a dataset of 5,85 billion CLIP-filtered image-text pairs, 14x bigger than LAION-400M, previously the biggest openly accessible image-text dataset in the world - see also our NeurIPS2022 paper Authors: Christoph Schuhmann, Richard Vencu, Romain Beaumont, Theo Coombes, Cade Gordon, Aarush K...
LAION-400-MILLION OPEN DATASET
by: Christoph Schuhmann, 20 Aug, 2021
We present LAION-400M: 400M English (image, text) pairs - see also our Data Centric AI NeurIPS Workshop 2021 paper Concept and Content The LAION-400M dataset is entirely openly, freely accessible. WARNING: be aware that this large-scale dataset is non-curated. It was built for research purposes to e...