# CLIP-ViT-L-14-laion2B-s32B-b82K by LAION eV: Open Model Source: https://savrn.com/models/clip-vit-l-14-laion2b-s32b-b82k Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- ## Runs On What it takes to serve CLIP-ViT-L-14-laion2B-s32B-b82K (428M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them. | Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits | | --- | --- | --- | --- | --- | --- | | 16-bit | 0.9 GB | 1.0 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 8-bit | 0.4 GB | 0.5 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 4-bit | 0.2 GB | 0.3 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the [SAVRN Index](https://savrn.com/ai-index/pricing/gpus), read Oct 24, 2026. [CLIP-ViT-L-14-laion2B-s32B-b82K on every accelerator the SAVRN Index prices, at every precision](https://savrn.com/models/clip-vit-l-14-laion2b-s32b-b82k/gpus) ## SAVRN's Notes on CLIP-ViT-L-14-laion2B-s32B-b82K Hand it a photo and a few label phrases you just wrote, and it tells you which phrase fits, with no training run; that is zero-shot image classification. The text side stops at 77 tokens, so labels are phrases, not paragraphs. At 16-bit the 428M parameters weigh 0.9 GB and the run needs 1.0 GB, a rounding error on the 192 GB MI300X we price at $1.85 an hour, so it shares a card rather than owning one. MIT permits commercial use, modification and redistribution with the notices kept, but weigh the publisher's framing too: LAION eV calls this a research output for research communities, trained on the LAION-2B English subset of LAION-5B with OpenCLIP. Before committing, run that data lineage through your governance review, check that English training text suits your labels, and decide whether a checkpoint released September 14, 2022 is current enough for your pipeline. SAVRN Research, 2026-09-18 ## Model Card By LAION eV, published under mit, revision 162703219714. ### Model Card for CLIP ViT-L/14 - LAION-2B ### Table of Contents 1. Model Details 2. Uses 3. Training Details 4. Evaluation 5. Acknowledgements 6. Citation 7. How To Get Started With the Model ### Model Details ### Model Description A CLIP ViT L/14 model trained with the LAION-2B English subset of LAION-5B (https://laion.ai/blog/laion-5b/) using OpenCLIP (https://github.com/mlfoundations/open_clip). Model training ('babysitting') done by Ross Wightman on the [JUWELS Booster](https://apps.fz-juelich.de/jsc/hps/juwels/booster-overview.html) supercomputer. See acknowledgements below. ### Uses As per the original [OpenAI CLIP model card](https://github.com/openai/CLIP/blob/d50d76daa670286dd6cacf3bcd80b5e4823fc8e1/model-card.md), 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 LAION-5B blog (https://laion.ai/blog/laion-5b/) and upcoming paper include additional discussion as it relates specifically to the training dataset. ### Direct Use [Read the full model card (1,396 words)](https://savrn.com/models/clip-vit-l-14-laion2b-s32b-b82k/card) ## Configuration Architecture CLIPModel Context length (tokens) 77 Layers 12 Hidden size 768 Feed-forward size 3,072 Attention heads 12 Vocabulary size 49,408 Stored precision float32 Model type clip ## Identity and Version Repository laion/CLIP-ViT-L-14-laion2B-s32B-b82K Publisher LAION eV Task Zero shot image classification Modality Other Library open_clip Parameters 428M parameters Languages Not stated by the source Revision 1627032197142fbe2a7cfec626f4ced3ae60d07a First published 2022-09-14 Last updated 2024-01-16 ## Files and Weights 31 files, 6.8 GB in total. The weights are 4 files totalling 6.8 GB in bin, safetensors. Weights4 files · 6.8 GB Configuration4 files · 5.8 KB Tokenizer4 files · 3.6 MB Documentation1 file · 11.4 KB Other17 files · 1.5 MB Repository1 file · 1.5 KB Every file | File | Type | Size | SHA-256 | | --- | --- | --- | --- | | model.safetensors | Weights | 1.7 GB | 0fce40ac6848 | | open_clip_pytorch_model.bin | Weights | 1.7 GB | 5ddb47339f44 | | open_clip_pytorch_model.safetensors | Weights | 1.7 GB | 7d129ed747e0 | | pytorch_model.bin | Weights | 1.7 GB | 45a6d8e04e46 | | config.json | Configuration | 4.6 KB | — | | open_clip_config.json | Configuration | 564 B | — | | preprocessor_config.json | Configuration | 275 B | — | | special_tokens_map.json | Configuration | 389 B | — | | README.md | Documentation | 11.4 KB | — | | runs/events.out.tfevents.1660090697.jwb0360.juwels.18885.0 | Other | 22.8 KB | bf2f8838ec8c | | runs/events.out.tfevents.1660150065.jwb0066.juwels.24737.0 | Other | 116.9 KB | 35d962fa7cd6 | | runs/events.out.tfevents.1660278820.jwb0066.juwels.20307.0 | Other | 112.4 KB | 21e58767af27 | | runs/events.out.tfevents.1660368981.jwb0069.juwels.12098.0 | Other | 115.4 KB | a2185da37e3b | | runs/events.out.tfevents.1660507060.jwb0038.juwels.22553.0 | Other | 114.4 KB | 294b98afa9a1 | | runs/events.out.tfevents.1660660016.jwb0075.juwels.30448.0 | Other | 112.8 KB | 015fbd1aaf18 | | runs/events.out.tfevents.1660757289.jwb0031.juwels.28174.0 | Other | 86.6 KB | b88d4d36282d | | runs/events.out.tfevents.1662171151.jwb0026.juwels.11856.0 | Other | 99.8 KB | ab81e18ee289 | | runs/events.out.tfevents.1662256342.jwb0043.juwels.17228.0 | Other | 97.3 KB | 366825801f79 | | runs/events.out.tfevents.1662354550.jwb0066.juwels.12897.0 | Other | 94.7 KB | 8bb8783d4ee1 | | runs/events.out.tfevents.1662437430.jwb0066.juwels.25458.0 | Other | 75.3 KB | cb64a95e3f8f | | runs/events.out.tfevents.1662584587.jwb0929.juwels.26132.0 | Other | 98.9 KB | 2b7c7b280a49 | | runs/events.out.tfevents.1662678523.jwb0577.juwels.25329.0 | Other | 96.3 KB | d965d500d6b8 | | runs/events.out.tfevents.1662777391.jwb0093.juwels.22511.0 | Other | 8.8 KB | 51f98864e746 | | runs/events.out.tfevents.1662784958.jwb0093.juwels.2143.0 | Other | 96.3 KB | a69693ecba0c | | runs/events.out.tfevents.1662923022.jwb0354.juwels.8816.0 | Other | 98.2 KB | 5f87516e52bc | | runs/events.out.tfevents.1663040222.jwb0024.juwels.20189.0 | Other | 8.1 KB | 7c4cbd6f93ad | | .gitattributes | Repository | 1.5 KB | — | | merges.txt | Tokenizer | 524.6 KB | — | | tokenizer.json | Tokenizer | 2.2 MB | — | | tokenizer_config.json | Tokenizer | 904 B | — | | vocab.json | Tokenizer | 862.3 KB | — | ## License and Download License mit Access Open weights, no gate Download size 6.8 GB [Download from LAION eV](https://huggingface.co/laion/CLIP-ViT-L-14-laion2B-s32B-b82K) Released by LAION eV through its official repository on Hugging Face. [Read the license](https://opensource.org/license/mit). ## Built From - Described by [arXiv:1910.04867](https://savrn.com/papers/a-large-scale-study-of-representation-learning-with-the-visual-task-adaptation-benchmark) - Described by arXiv:2110.09456 - Described by [arXiv:2111.09883](https://savrn.com/papers/swin-transformer-v2-scaling-up-capacity-and-resolution) ## Memory Requirements | Precision | Weights in memory | | --- | --- | | As published | 6.8 GB | | 16-bit | 0.9 GB | | 8-bit | 0.4 GB | | 4-bit | 0.2 GB | Weights only, from the published parameter count; the key-value cache and runtime add to this. ## Compare CLIP-ViT-L-14-laion2B-s32B-b82K - [CLIP-ViT-L-14-laion2B-s32B-b82K vs clip-vit-large-patch14](https://savrn.com/models/compare/clip-vit-l-14-laion2b-s32b-b82k-vs-clip-vit-large-patch14) - [CLIP-ViT-L-14-laion2B-s32B-b82K vs siglip2-base-patch16-224](https://savrn.com/models/compare/clip-vit-l-14-laion2b-s32b-b82k-vs-siglip2-base-patch16-224) - [CLIP-ViT-L-14-laion2B-s32B-b82K vs siglip2-base-patch16-naflex](https://savrn.com/models/compare/clip-vit-l-14-laion2b-s32b-b82k-vs-siglip2-base-patch16-naflex) ## Questions About CLIP-ViT-L-14-laion2B-s32B-b82K ### How much GPU memory does CLIP-ViT-L-14-laion2B-s32B-b82K need? About 1 GB at 16-bit and 0.3 GB at 4-bit: the weights (428M parameters) plus a working margin. A long context needs more. ### What is the cheapest GPU to run CLIP-ViT-L-14-laion2B-s32B-b82K on? At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists. ### Can I use CLIP-ViT-L-14-laion2B-s32B-b82K commercially? Yes. CLIP-ViT-L-14-laion2B-s32B-b82K is released under MIT License. The MIT License is a short permissive license. It permits commercial use, modification and redistribution, provided the copyright notice and permission notice are included. ### What is CLIP-ViT-L-14-laion2B-s32B-b82K's context length? 77 tokens, from the maximum position embeddings in its published configuration. ## Similar Models Model · Zero shot image classification ### [clip-vit-large-patch14](https://savrn.com/models/clip-vit-large-patch14) [OpenAI](https://savrn.com/model-publishers/openai) Disclaimer: The model card is taken and modified from the official CLIP repository, it can be found here. The CLIP model was developed by researchers at OpenAI to learn about what contributes to robustness in computer vision tasks. The model was also developed to test the ability of models to generalize to arbitrary image classification tasks in a zero-shot manner. It was not developed for general model deployment - to deploy models like CLIP, researchers will first need to carefully study their capabilities in relation to the specific context they’re being deployed within. January 2021 The base model uses a ViT-L/14 Transformer architecture as an image encoder and uses a masked… Open weights 428M parameters 77 tokens transformers [View model](https://savrn.com/models/clip-vit-large-patch14) Model · Zero shot image classification ### [siglip2-base-patch16-256](https://savrn.com/models/siglip2-base-patch16-256) [Google](https://savrn.com/model-publishers/google) SigLIP 2 extends the pretraining objective of SigLIP with prior, independently developed techniques into a unified recipe, for improved semantic understanding, localization, and dense features. You can use the raw model for tasks like zero-shot image classification and image-text retrieval, or as a vision encoder for VLMs (and other vision tasks). Here is how to use this model to perform zero-shot image classification: You can encode an image using the Vision Tower like so: For more code examples, we refer to the siglip documentation. SigLIP 2 adds some clever training objectives on top of SigLIP: SigLIP 2 is pre-trained on the WebLI dataset (Chen et al., 2023). The model was trained on up… Open weights apache-2.0 375M parameters transformers [View model](https://savrn.com/models/siglip2-base-patch16-256) Model · Zero shot image classification ### [siglip2-base-patch16-naflex](https://savrn.com/models/siglip2-base-patch16-naflex) [Google](https://savrn.com/model-publishers/google) SigLIP 2 extends the pretraining objective of SigLIP with prior, independently developed techniques into a unified recipe, for improved semantic understanding, localization, and dense features. You can use the raw model for tasks like zero-shot image classification and image-text retrieval, or as a vision encoder for VLMs (and other vision tasks). Here is how to use this model to perform zero-shot image classification: You can encode an image using the Vision Tower like so: For more code examples, we refer to the siglip2 documentation. SigLIP 2 adds some clever training objectives on top of SigLIP: SigLIP 2 is pre-trained on the WebLI dataset (Chen et al., 2023). The model was trained on up… Open weights apache-2.0 375M parameters transformers [View model](https://savrn.com/models/siglip2-base-patch16-naflex) Model · Zero shot image classification ### [siglip2-base-patch16-224](https://savrn.com/models/siglip2-base-patch16-224) [Google](https://savrn.com/model-publishers/google) SigLIP 2 extends the pretraining objective of SigLIP with prior, independently developed techniques into a unified recipe, for improved semantic understanding, localization, and dense features. You can use the raw model for tasks like zero-shot image classification and image-text retrieval, or as a vision encoder for VLMs (and other vision tasks). Here is how to use this model to perform zero-shot image classification: You can encode an image using the Vision Tower like so: For more code examples, we refer to the siglip documentation. SigLIP 2 adds some clever training objectives on top of SigLIP: SigLIP 2 is pre-trained on the WebLI dataset (Chen et al., 2023). The model was trained on up… Open weights apache-2.0 375M parameters transformers [View model](https://savrn.com/models/siglip2-base-patch16-224) Model · Zero shot image classification ### [siglip-base-patch16-224](https://savrn.com/models/siglip-base-patch16-224) [Google](https://savrn.com/model-publishers/google) SigLIP model pre-trained on WebLi at resolution 224x224. It was introduced in the paper Sigmoid Loss for Language Image Pre-Training by Zhai et al. and first released in this repository. Disclaimer: The team releasing SigLIP did not write a model card for this model so this model card has been written by the Hugging Face team. SigLIP is CLIP, a multimodal model, with a better loss function. The sigmoid loss operates solely on image-text pairs and does not require a global view of the pairwise similarities for normalization. This allows further scaling up the batch size, while also performing better at smaller batch sizes. A TLDR of SigLIP by one of the authors can be found here. You can use… Open weights apache-2.0 203M parameters transformers [View model](https://savrn.com/models/siglip-base-patch16-224) Model · Zero shot image classification ### [CLIP-ViT-B-32-laion2B-s34B-b79K](https://savrn.com/models/clip-vit-b-32-laion2b-s34b-b79k) [LAION eV](https://savrn.com/model-publishers/laion) A CLIP ViT-B/32 model trained with the LAION-2B English subset of LAION-5B (https://laion.ai/blog/laion-5b/) using OpenCLIP (https://github.com/mlfoundations/openclip). Model training done by Romain Beaumont on the stability.ai cluster. 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.… Open weights mit 151M parameters 77 tokens open_clip [View model](https://savrn.com/models/clip-vit-b-32-laion2b-s34b-b79k) ## LAION eV [All models and datasets](https://savrn.com/model-publishers/laion) ## Versions - [162703219714](https://savrn.com/models/clip-vit-l-14-laion2b-s32b-b82k/versions/162703219714) · current 2026-10-24 ## Explore More - [All zero shot image classification models](https://savrn.com/models/tasks/zero-shot-image-classification) - [All models under mit](https://savrn.com/models/licenses/mit) - [Model comparisons](https://savrn.com/models/comparisons) - [The model directory](https://savrn.com/models) - [Open model prices by host](https://savrn.com/ai-index/pricing/open-models) ## Source - Repository metadata, read 2026-10-19. - [LAION eV website](https://laion.ai/) - [Hugging Face record](https://huggingface.co/laion/CLIP-ViT-L-14-laion2B-s32B-b82K) - [How the hub is built](https://savrn.com/model-hub/methodology)